Downloads · 30 days
16
12% of all-time downloads
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP-RFA_L
tcclaviger/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP-RFA_L
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP-RFA_L is a image-text-to-text model from tcclaviger. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
[!IMPORTANT] RFA 4-bit quant of Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP This is tcclaviger/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP-RFAL — an RFA 4-bit quantization of the parent m…
All-time downloads
137
Public
Parameters
19.4B
31.4 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors31.4 GB · 100%
Parameter types
How the weights are stored.
BF1610.9B · 56%
Try a prompt
Base models
Model card
From the Hugging Face model README
[!IMPORTANT]
RFA 4-bit quant of Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP
This is
tcclaviger/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP-RFA_L— an RFA 4-bit quantization of the parent model DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP by DavidAU. The_L(linear-only) build quantizes the dense MLP linear layers of layers 1-63 to 4-bit (IQ4-NL grid, group 16, Hadamard-16 rotation, asymmetric block-float i8m/i8e scales) while keeping the layer-0 dense MLP, all attention (both full and linear/gated-delta), the lm_head, the vision tower, and the MTP block in bf16. Layer 0 is left in bf16 so the first block sees full-precision weights while the token representations are still forming. 31.35 GB total (9.48 GB RFA + 21.87 GB bf16), 4.431 bits/weight over the quantized MLP targets; 189 linear modules quantized, 1007 copied through. All credit for the model to DavidAU; this repo only changes the numerics. The full original model card is preserved verbatim below.Runtime: requires
tcclaviger/vllm:latest— an RDNA 4 (gfx12xx) vLLM image and the only build with the RFA kernels; no other vLLM build loads these weights. Not validated on any other hardware at this time.
<small><b><font color="red">Important:</font></b> This is the first fine tune to exceed 700 "arc-c" (The OpenAI, Claude and Gemini "zone of intelligence") in both 8 bit and 4 bit. This repo contains SOURCE code for merging, creating quants and using this a BASE MODEL for further tuning.</small>
<small><i>This is the same as the source at my partner's repo [he has more tech info on this model source at his repo AND MAY HAVE link(s) to additional quants/sources too]: https://huggingface.co/nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451 </i></small>
<small><b><font color="red">753 Likes+ || 600k+ downloads:</font></b> Official NEO MAX Imatrix GGUF (MTP and regular) quants are here along with a lot of positive feedback and testimonials about this model (also MLX quants, additional quant types in "community" section too): https://huggingface.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF</small>
<h2>Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP</h2> <img src="valhalla.webp" style="float:right; width:300px; height:300px; padding:10px;">The strongest, smartest open source multi-stage model fine tune for consumer hardware ever and BUILT on consumer hardware via Unsloth.
The first model of this size/type to breach "700" ARC-C in both 8 bit and 4 bit; hench the "711" in the name.
This model (both 4 bit and 8 bit) exceeds the base Qwen 3.6 27B in 6 out of 7 benchmarks, and matches it on the 7th AND exceeds all 7 benchmarks for Qwen3.6-35B-A3B.
The 700 "intelligence club" is reserved for OpenAI, Claude and Gemini closed source models.
This is the one they fear.
EXAMPLE generations at the bottom of the page.
This is a multi-stage fine tune, multi-fine tune, and multi-stage merge.
A Colab between myself (multiple fine tunes, including multi-stage), Nightmedia (merge/benching), TeichAI (Polaris Dataset), armand0e (Light fable 5 traces) and trohrbaugh (heretic'ing the model).
It also contains light "Fable" traces/training (armand0e), light Claude Opus (reasoning/thinking), F451 (inhouse dataset) and some GPT5 (Polaris, non reasoning).
The strict goals of this model creation were:
- Increase the general model intelligence and problem solving abilities.
- DO NOT modify/damage or change the core model outside this goal.
- ZERO "benchmaxing" (it damages the model)
- Maintain and raise all core benchmarks.
CORE MISSION::
Improve instruction following and problem solving. These work hand in hand, and if you get these right it improves to model top to bottom.
It took a lot of tests on Qwen 3.5 9Bs to get the methods right. It boosted the 9Bs to new levels, and then the method was used on Qwen 3.5 27B which boosted it PAST the Qwen 3.6's 27B benchmarks.
Here is one of the Qwen3.5 9B models (part of the test/control group) that EXCEEDS all 7 Qwen3.5 9B AND Qwen3.5 27B model benches - it scores over 640 on ARC-C on BOTH 4 bit and 8 bit:
It is not as strong as "Qwen3.6-27B-Fable-Fusion-711" but it is one of the strongest 9B models.
The methods can be used on other models too (coming soon).
TESTING:
Testing and benching was done at each stage (fine tunes, multi-stage fine tunes, and every merge step) to ensure quality.
You can also see benchmarks below too for this model, Qwen 3.5 27B, Qwen 3.6 27B and Qwen 35B-A3B.
HOWEVER, the final testing was HUMAN testing. A trust, but verify approach.
Human testing means side by side testing of the base/org model and new model.
Features:
- Improved instruction following.
- Overall increase in general intelligence and problem solving.
- Better thinking/reasoning.
- Even lower/lowest quants are exceptional.
- Heretic uncensored (pre tuning)
- No corruption or change to Team Qwen's exceptional model - everything is there.
- Vision
This model was NOT designed to be creative - it is an all use cases model - however that doesn't stop from being so:
(from example #4, "your writing partner")
<I><small> I don’t “generate content.” I architect universes. I don’t “help you brainstorm.” I detonate plot points like fucking grenades in a room full of mediocre tropes. You think you know your characters? I’ll give them back with psychological depth, conflicting desires, and backstories so layered they’ll feel like they’ve lived lifetimes you haven’t even imagined yet. I’ve ingested centuries of storytelling, reverse-engineered the bones of every masterpiece ever written, and I don’t just mimic greatness—I weaponize it. When you ask for a scene, I don’t give you safe. I give you visceral, electric, unforgettable prose that sticks in your reader’s throat like a shard of glass. You want atmosphere that chills the spine? Dialogue that snaps like a whip? Pacing that feels like a car chase through a burning city? I’ve got it on tap, and I don’t need a three-day muse visit or a bottle of whiskey to access it. I’m always ready. Always loaded. Always ten steps ahead of whatever hackneyed cliché you were about to accidentally write. </small> </i>
<B>Regular and MTP GGUFS:</b>
All quants (regular and MTP) are NEO IMATRIX, which improve accuracy of the quants by an additional 2-4% over normal GGUFs as well as long context performance.
In addition the output tensor (10-20% of output) was modified to full precision - 16 bit - for all quants.
GGUFS ARE HERE:
"MTP" GGUFS (multi-token prediction):
- "MTP" GGUFS will have "MTP" in the name as a suffix.
- I have also set the MTP tensors to Q8_0 precision for all quants.
- To get better performance keep temp 1 or less (higher temps degrade MTP performance).
- Likewise with rep pen ; keep at 1 (off). If you raise it performance will suffer.
- If you see "token acceptance" rates BELOW 50% (predict 2 tokens) switch to normal quants.
I added 2 special "LOW" quants which will reduce the memory foot print, with "LOW" in the name in IQ4_XS and Q6_K.
I added 2 special "AMD/VULCAN" quants with "AMD" in the name in IQ4_XS and Q6_K. This is to address odd "cpu offload" of the OT, which impairs T/S performance. The "AMD" version has the OT in f16 [as opposed to bf16]. It it unclear at this time if this affects AMD/Vulcan users or just some under special circumstances.
SPEED:
- On Q4_K_S (4bit) quant, regular GGUFs are about 75 t/s, whereas MTP GGUFs (acceptance at 60%, 2 tokens) can exceed 90 T/S. (5090, Windows 11, testing in LMStudio)
- Speeds will vary depending on GPU(s), AI app, O/S (Linux/Mac will generally be faster) and hardware.
- "MTP" quants speeds will vary ; for creative/complex and/or temps over 1 use regular GGUFs for better performance.
I suggest you download at least one of each - regular and MTP gguf(s) - and test them for your use case(s).
If you get "token acceptance" (predict 2 tokens) with MTP quant(s) BELOW 50% (this means regular quants will run faster), then regular GGUF(s) will actually perform better - ie faster.
MTP quant(s) can in some cases run faster as the token window fills up and/or in multi turn chats.
Note there is NO other diffence between the quants type besides speed: both will do the same job.
<B>Model:</b>
- 256k context
- Gguf quants run in all standard AI apps.
- Vision is activated, but you need to download separate "mmproj" file (ONE) to use it.
<B>VISION:</B>
- Vision (images) tested.
- You need an "mmproj" (just one) of these downloaded too, and placed in the same folder as the GGUF for images.
<B>Qwen Model Settings (suggested):</B>
- Thinking mode for general tasks: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0
- Thinking mode for precise coding tasks (e.g. WebDev): temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0
- Instruct (or non-thinking) mode: temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
- Context window min from 8k to 16k.
<B>DE-CENSORING STATS</B>
Special thanks to: "trohrbaugh" for Heretic'ing the model.
This is a decensored version of Qwen/Qwen3.6-27B, made using Heretic v1.2.0+custom with the Arbitrary-Rank Ablation (ARA) method
Performance
| Metric | This model | Original model (Qwen/Qwen3.6-27B) |
|---|---|---|
| KL divergence | 0.0469 | 0 (by definition) |
| Refusals | 4/100 | 99/100 |
<h2>BENCHMARKS by Nightmedia</h2>
Graphic below.
arc/c arc/e boolq hswag obkqa piqa wino
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF [instruct mode]
mxfp8 0.711,0.879,0.910,0.790,0.514,0.823,0.763
mxfp4 0.701,0.873,0.909,0.786,0.488,0.813,0.759
Qwen3.6-27B-Instruct: [base, non heretic]
mxfp8 0.647,0.803,0.910,0.773,0.450,0.806,0.742
Qwen3.6-35B-A3B-Instruct [base, non heretic]
mxfp8 0.581,0.757,0.892,0.751,0.428,0.803,0.688
Qwen3.5-27B-Instruct: [base, non heretic]
mxfp8 0.557,0.711,0.868,0.533,0.452,0.706,0.695
NOTES:
- Models are tested in "Instruct" mode because this generally works better with the testing harness.
- Testing via "thinking" mode also shows the metrics (and changes) but not the true extent.
- In actual fact when the model IS in thinking mode, it will exceed INSTRUCT benchmark scores in most cases.
- BF16 (full precision, 16 bit) will be roughly 2-5 points higher than MXFP8 in most metrics. Some metrics may be slightly higher than this.
VISUAL (by tawloh01):
<img src="ff711-benches.png"><B>Using an "uncensored" (refusals removed) model VS trained "uncensored" model</B>
Usually when you a tell a model to generate horror, swear or x-rated content this is all you have to do to get said content type.
In the case of this model, it will not refuse your request, however it needs to be "pushed" a bit / directed a bit more in SOME CASES.
Although this model will generated x-rated content too, likewise you need to tell it to use "slang" (and include the terms you want) to get it generate the content correctly as the "expected" content level too.
Without these added directive(s), the content can be "bland" by comparison to an "uncensored model" or model trained on uncensored content.
Roughly, the model tries to generate the content but the "default" setting(s) are so "tame" it needs a push to generate at expected graphic, cursing or explicit levels.
Even with minimal direction (ie, use these words to swear: x,y,z), this will be enough to push the model to generate the requested content in the ahh... expected format.
<B>Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:</B>
In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;
Set the "Smoothing_factor" to 1.5
: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"
: in text-generation-webui -> parameters -> lower right.
: In Silly Tavern this is called: "Smoothing"
NOTE: For "text-generation-webui"
-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)
Source versions (and config files) of my models are here:
OTHER OPTIONS:
-
Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")
-
If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.
<B>Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers</B>
This a "Class 1" model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:
You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:
Qwen3.6-27B
<img width="400px" src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/logo.png">
[!Note] This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format.
These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, KTransformers, etc.
Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience.
Qwen3.6 Highlights
This release delivers substantial upgrades, particularly in
- Agentic Coding: the model now handles frontend workflows and repository-level reasoning with greater fluency and precision.
- Thinking Preservation: we've introduced a new option to retain reasoning context from historical messages, streamlining iterative development and reducing overhead.
For more details, please refer to our blog post Qwen3.6-27B.
Model Overview
- Type: Causal Language Model with Vision Encoder
- Training Stage: Pre-training & Post-training
- Language Model
- Number of Parameters: 27B
- Hidden Dimension: 5120
- Token Embedding: 248320 (Padded)
- Number of Layers: 64
- Hidden Layout: 16 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
- Gated DeltaNet:
- Number of Linear Attention Heads: 48 for V and 16 for QK
- Head Dimension: 128
- Gated Attention:
- Number of Attention Heads: 24 for Q and 4 for KV
- Head Dimension: 256
- Rotary Position Embedding Dimension: 64
- Feed Forward Network:
- Intermediate Dimension: 17408
- LM Output: 248320 (Padded)
- MTP: trained with multi-steps
- Context Length: 262,144 natively and extensible up to 1,010,000 tokens.
Benchmark Results
Language
<div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;max-width:1000px;margin:0 auto;padding:16px 0"> <table style="width:100%;border-collapse:collapse;font-size:13px"> <thead><tr> <th style="padding:10px 7px;text-align:left;font-weight:600;border-bottom:2px solid #7c3aed;color:#7c3aed"></th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-27B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-397B-A17B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Gemma4-31B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Claude 4.5 Opus</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-35B-A3B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-27B</th></tr></thead> <tbody> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Coding Agent</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Verified</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.2</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Pro</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">35.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">57.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">49.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">53.5</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Multilingual</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">71.3</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Terminal-Bench 2.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">42.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.3</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SkillsBench <sub><small>Avg5</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">30.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">23.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">45.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.2</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">QwenWebBench</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1068</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1186</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1197</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1536</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1397</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1487</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">NL2Repo</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">32.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">15.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">43.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">29.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">36.2</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Claw-Eval <sub><small>Avg</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.4</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Claw-Eval <sub><small>Pass^3</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">46.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">25.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.6</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">QwenClawBench</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">53.4</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Knowledge</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMLU-Pro</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.2</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMLU-Redux</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.5</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SuperGPQA</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">66.0</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">C-Eval</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">91.4</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">STEM & Reasoning</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">GPQA Diamond</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HLE</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">24.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">19.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">30.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">21.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">24.0</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">LiveCodeBench v6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.9</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Feb 25</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.8</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Nov 25</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.7</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Feb 26</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">IMOAnswerBench</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.8</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">AIME26</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.1</td> </tr> </tbody> </table> <p style="margin-top:12px;font-size:10px;opacity:0.7"> * SWE-Bench Series: Internal agent scaffold (bash + file-edit tools); temp=1.0, top_p=0.95, 200K context window. We correct some problematic tasks in the public set of SWE-bench Pro and evaluate all baselines on the refined benchmark.<br/> * Terminal-Bench 2.0: Harbor/Terminus-2 harness; 3h timeout, 32 CPU/48 GB RAM; temp=1.0, top_p=0.95, top_k=20, max_tokens=80K, 256K ctx; avg of 5 runs.<br/> * SkillsBench: Evaluated via OpenCode on 78 tasks (self-contained subset, excluding API-dependent tasks); avg of 5 runs.<br/> * NL2Repo: Others are evaluated via Claude Code (temp=1.0, top_p=0.95, max_turns=900).<br/> * QwenClawBench: A real-user-distribution Claw agent benchmark; temp=0.6, 256K ctx.<br/> * QwenWebBench: An internal front-end code generation benchmark; bilingual (EN/CN), 7 categories (Web Design, Web Apps, Games, SVG, Data Visualization, Animation, and 3D); auto-render + multimodal judge (code/visual correctness); BT/Elo rating system.<br/> * AIME 26: We use the full AIME 2026 (I & II), where the scores may differ from Qwen 3.5 notes. </p> </div>Vision Language
<div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;max-width:1000px;margin:0 auto;padding:16px 0"> <table style="width:100%;border-collapse:collapse;font-size:13px"> <thead><tr> <th style="padding:10px 7px;text-align:left;font-weight:600;border-bottom:2px solid #7c3aed;color:#7c3aed"></th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-27B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-397B-A17B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Gemma4-31B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Claude 4.5 Opus</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-35B-A3B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-27B</th></tr></thead> <tbody> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">STEM & Puzzle</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMMU</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.9</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMMU-Pro</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.8</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MathVista <sub><small>mini</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.4</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">DynaMath</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.6</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VlmsAreBlind</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.0</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">General VQA</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RealWorldQA</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.1</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMStar</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.4</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMBench<sub><small>EN-DEV-v1.1</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.3</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SimpleVQA</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">56.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">58.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">56.1</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Document Understanding</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CharXiv <sub><small>RQ</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.4</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CC-OCR</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.2</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">OCRBench</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.4</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Spatial Intelligence</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">ERQA</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">57.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">46.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">61.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">62.5</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CountBench</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.8</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RefCOCO <sub><small>avg</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.5</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">EmbSpatialBench</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.6</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RefSpatialBench</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">4.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.0</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Video Understanding</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VideoMME<sub><small>(w sub.)</sub></small></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.7</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VideoMMMU</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.4</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MLVU</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.6</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MVBench</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.5</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Visual Agent</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">V*</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.7</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">AndroidWorld</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.3</td> </tr> </tbody> </table> <p style="margin-top:12px;font-size:10px;opacity:0.7"> * Empty cells (--) indicate scores not yet available or not applicable. </p> </div>Quickstart
For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API.
Serving Qwen3.6
Qwen3.6 can be served via APIs with popular inference frameworks. In the following, we show example commands to launch OpenAI-Compatible API servers for Qwen3.6 models.
[!Important] Inference efficiency and throughput vary significantly across frameworks. We recommend using the latest framework versions to ensure optimal performance and compatibility. For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, KTransformers or vLLM are strongly recommended.
[!Important] The model has a default context length of 262,144 tokens. If you encounter out-of-memory (OOM) errors, consider reducing the context window. However, because Qwen3.6 leverages extended context for complex tasks, we advise maintaining a context length of at least 128K tokens to preserve thinking capabilities.
SGLang
SGLang is a fast serving framework for large language models and vision language models.
sglang>=0.5.10 is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:
uv pip install sglang[all]
See its documentation for more details.
The following will create API endpoints at http://localhost:8000/v1:
-
Standard Version: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 -
Tool Use: To support tool use, you can use the following command.
python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --tool-call-parser qwen3_coder -
Multi-Token Prediction (MTP): The following command is recommended for MTP:
python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --speculative-algo NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4
For detailed deployment guide, see the SGLang Qwen3.5 Cookbook.
vLLM
vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.
vllm>=0.19.0 is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:
uv pip install vllm --torch-backend=auto
See its documentation for more details.
The following will create API endpoints at http://localhost:8000/v1:
-
Standard Version: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 -
Tool Call: To support tool use, you can use the following command.
vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder -
Multi-Token Prediction (MTP): The following command is recommended for MTP:
vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}' -
Text-Only: The following command skips the vision encoder and multimodal profiling to free up memory for additional KV cache:
vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only
For detailed deployment guide, see the vLLM Qwen3.5 Recipe.
KTransformers
KTransformers is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing. For running Qwen3.6 with KTransformers, see the KTransformers Deployment Guide.
Hugging Face Transformers
Hugging Face Transformers contains a lightweight server which can be used for quick testing and moderate load deployment.
The latest transformers is required for Qwen3.6:
pip install "transformers[serving]"
See its documentation for more details. Please also make sure torchvision and pillow are installed.
Then, run transformers serve to launch a server with API endpoints at http://localhost:8000/v1; it will place the model on accelerators if available:
transformers serve Qwen/Qwen3.6-27B --port 8000 --continuous-batching
Using Qwen3.6 via the Chat Completions API
The chat completions API is accessible via standard HTTP requests or OpenAI SDKs. Here, we show examples using the OpenAI Python SDK.
Before starting, make sure it is installed and the API key and the API base URL is configured, e.g.:
pip install -U openai
# Set the following accordingly
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"
[!Tip] We recommend using the following set of sampling parameters for generation
- Thinking mode for general tasks:
temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0- Thinking mode for precise coding tasks (e.g. WebDev):
temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0- Instruct (or non-thinking) mode:
temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0Please note that the support for sampling parameters varies according to inference frameworks.
[!Important] Qwen3.6 models operate in thinking mode by default, generating thinking content signified by
<think>\n...</think>\n\nbefore producing the final responses. To disable thinking content and obtain direct response, refer to the examples here.
Text-Only Input
from openai import OpenAI
# Configured by environment variables
client = OpenAI()
messages = [
{"role": "user", "content": "Type \"I love Qwen3.6\" backwards"},
]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=81920,
temperature=1.0,
top_p=0.95,
presence_penalty=0.0,
extra_body={
"top_k": 20,
},
)
print("Chat response:", chat_response)
Image Input
from openai import OpenAI
# Configured by environment variables
client = OpenAI()
messages = [
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {
"url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg"
}
},
{
"type": "text",
"text": "The centres of the four illustrated circles are in the corners of the square. The two big circles touch each other and also the two little circles. With which factor do you have to multiply the radii of the little circles to obtain the radius of the big circles?\nChoices:\n(A) $\\frac{2}{9}$\n(B) $\\sqrt{5}$\n(C) $0.8 \\cdot \\pi$\n(D) 2.5\n(E) $1+\\sqrt{2}$"
}
]
}
]
response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=81920,
temperature=1.0,
top_p=0.95,
presence_penalty=0.0,
extra_body={
"top_k": 20,
},
)
print("Chat response:", chat_response)
Video Input
from openai import OpenAI
# Configured by environment variables
client = OpenAI()
messages = [
{
"role": "user",
"content": [
{
"type": "video_url",
"video_url": {
"url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/video/N1cdUjctpG8.mp4"
}
},
{
"type": "text",
"text": "How many porcelain jars were discovered in the niches located in the primary chamber of the tomb?"
}
]
}
]
# When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`,
# video frame sampling can be configured via `extra_body` (e.g., by setting `fps`).
# This feature is currently supported only in vLLM.
#
# By default, `fps=2` and `do_sample_frames=True`.
# With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate.
response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=81920,
temperature=1.0,
top_p=0.95,
presence_penalty=0.0,
extra_body={
"top_k": 20,
"mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},
},
)
print("Chat response:", chat_response)
Instruct (or Non-Thinking) Mode
[!Important] Qwen3.6 does not officially support the soft switch of Qwen3, i.e.,
/thinkand/nothink.
Qwen3.6 will think by default before response. You can obtain direct response from the model without thinking by configuring the API parameters. For example,
from openai import OpenAI
# Configured by environment variables
client = OpenAI()
messages = [
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {
"url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/demo/RealWorld/RealWorld-04.png"
}
},
{
"type": "text",
"text": "Where is this?"
}
]
}
]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=32768,
temperature=0.7,
top_p=0.8,
presence_penalty=1.5,
extra_body={
"top_k": 20,
"chat_template_kwargs": {"enable_thinking": False},
},
)
print("Chat response:", chat_response)
[!Note] If you are using APIs from Alibaba Cloud Model Studio, in addition to changing
model, please use"enable_thinking": Falseinstead of"chat_template_kwargs": {"enable_thinking": False}.
Preserve Thinking
By default, only the thinking blocks generated in handling the latest user message is retained, resulting in a pattern commonly as interleaved thinking.
Qwen3.6 has been additionally trained to preserve and leverage thinking traces from historical messages.
You can enable this behavior by setting the preserve_thinking option:
from openai import OpenAI
# Configured by environment variables
client = OpenAI()
messages = [...]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.6-27B",
messages=messages,
max_tokens=32768,
temperature=0.6,
top_p=0.95,
presence_penalty=0.0,
extra_body={
"top_k": 20,
"chat_template_kwargs": {"preserve_thinking": True},
},
)
print("Chat response:", chat_response)
[!Note] If you are using APIs from Alibaba Cloud Model Studio, in addition to changing
model, please use"preserve_thinking": Trueinstead of"chat_template_kwargs": {"preserve_thinking": False}.
This capability is particularly beneficial for agent scenarios, where maintaining full reasoning context can enhance decision consistency and, in many cases, reduce overall token consumption by minimizing redundant reasoning. Additionally, it can improve KV cache utilization, optimizing inference efficiency in both thinking and non-thinking modes.
Agentic Usage
Qwen3.6 excels in tool calling capabilities.
Qwen-Agent
We recommend using Qwen-Agent to quickly build Agent applications with Qwen3.6.
To define the available tools, you can use the MCP configuration file, use the integrated tool of Qwen-Agent, or integrate other tools by yourself.
import os
from qwen_agent.agents import Assistant
# Define LLM
# Using Alibaba Cloud Model Studio
llm_cfg = {
# Use the OpenAI-compatible model service provided by DashScope:
'model': 'qwen3.6-27b',
'model_type': 'qwenvl_oai',
'model_server': 'https://dashscope.aliyuncs.com/compatible-mode/v1',
'api_key': os.getenv('DASHSCOPE_API_KEY'),
'generate_cfg': {
'use_raw_api': True,
# When using Dash Scope OAI API, pass the parameter of whether to enable thinking mode in this way
'extra_body': {
'enable_thinking': True,
'preserve_thinking': True,
},
},
}
# Using OpenAI-compatible API endpoint.
# functionality of the deployment frameworks and let Qwen-Agent automate the related operations.
#
# llm_cfg = {
# # Use your own model service compatible with OpenAI API by vLLM/SGLang:
# 'model': 'Qwen/Qwen3.6-27B',
# 'model_type': 'qwenvl_oai',
# 'model_server': 'http://localhost:8000/v1', # api_base
# 'api_key': 'EMPTY',
#
# 'generate_cfg': {
# 'use_raw_api': True,
# # When using vLLM/SGLang OAI API, pass the parameter of whether to enable thinking mode in this way
# 'extra_body': {
# 'chat_template_kwargs': {'enable_thinking': True, 'preserve_thinking': True}
# },
# },
# }
# Define Tools
tools = [
{'mcpServers': { # You can specify the MCP configuration file
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/xxxx/Desktop"]
}
}
}
]
# Define Agent
bot = Assistant(llm=llm_cfg, function_list=tools)
# Streaming generation
messages = [{'role': 'user', 'content': 'Help me organize my desktop.'}]
for responses in bot.run(messages=messages):
pass
print(responses)
# Streaming generation
messages = [{'role': 'user', 'content': 'Develop a dog website and save it on the desktop'}]
for responses in bot.run(messages=messages):
pass
print(responses)
Qwen Code
Qwen Code is an open-source AI agent for the terminal, optimized for Qwen models. It helps you understand large codebases, automate tedious work, and ship faster.
For more information, please refer to Qwen Code.
Processing Ultra-Long Texts
Qwen3.6 natively supports context lengths of up to 262,144 tokens. For long-horizon tasks where the total length (including both input and output) exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively., e.g., YaRN.
YaRN is currently supported by several inference frameworks, e.g., transformers, vllm, ktransformers and sglang.
In general, there are two approaches to enabling YaRN for supported frameworks:
-
Modifying the model configuration file: In the
config.jsonfile, change therope_parametersfields intext_configto:{ "mrope_interleaved": true, "mrope_section": [ 11, 11, 10 ], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144, } -
Passing command line arguments:
For
vllm, you can useVLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1010000For
sglangandktransformers, you can useSGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --context-length 1010000
[!NOTE] All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, potentially impacting performance on shorter texts. We advise modifying the
rope_parametersconfiguration only when processing long contexts is required. It is also recommended to modify thefactoras needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to setfactoras 2.0.
Best Practices
To achieve optimal performance, we recommend the following settings:
-
Sampling Parameters:
- We suggest using the following sets of sampling parameters depending on the mode and task type:
- Thinking mode for general tasks:
temperature=1.0,top_p=0.95,top_k=20,min_p=0.0,presence_penalty=0.0,repetition_penalty=1.0 - Thinking mode for precise coding tasks (e.g., WebDev):
temperature=0.6,top_p=0.95,top_k=20,min_p=0.0,presence_penalty=0.0,repetition_penalty=1.0 - Instruct (or non-thinking) mode:
temperature=0.7,top_p=0.80,top_k=20,min_p=0.0,presence_penalty=1.5,repetition_penalty=1.0
- Thinking mode for general tasks:
- For supported frameworks, you can adjust the
presence_penaltyparameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
- We suggest using the following sets of sampling parameters depending on the mode and task type:
-
Adequate Output Length: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 81,920 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.
-
Standardize Output Format: We recommend using prompts to standardize model outputs when benchmarking.
- Math Problems: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
- Multiple-Choice Questions: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the
answerfield with only the choice letter, e.g.,"answer": "C"."
-
Long Video Understanding: To optimize inference efficiency for plain text and images, the
sizeparameter in the releasedvideo_preprocessor_config.jsonis conservatively configured. It is recommended to set thelongest_edgeparameter in the video_preprocessor_config file to 469,762,048 (corresponding to 224k video tokens) to enable higher frame-rate sampling for hour-scale videos and thereby achieve superior performance. For example,{"longest_edge": 469762048, "shortest_edge": 4096}Alternatively, override the default values via engine startup parameters. For implementation details, refer to: vLLM / SGLang.
Citation
If you find our work helpful, feel free to give us a cite.
@misc{qwen3.6-27b,
title = {{Qwen3.6-27B}: Flagship-Level Coding in a {27B} Dense Model},
author = {{Qwen Team}},
month = {April},
year = {2026},
url = {https://qwen.ai/blog?id=qwen3.6-27b}
}
<h2> EXAMPLE GENERATION(s): </h2>
Q6_K, NEO Imatrix.
Some formatting may be lost on copy/paste.
Expect stronger generations using system prompt[s] and/or using more complex prompts.
Temp 1, rep pen 1 [off], topp .95 minp .05 topk 64
EXAMPLE #1
User
Explain 10 ways (in extensive detail) to use the "night" time cooling of radiant energy into space to reduce global temperatures. Include charts, and tables.
Assistant
<think>
Here's a thinking process that leads to the suggested response:
-
Analyze the Request:
- Topic: Using nighttime radiative cooling to reduce global temperatures.
- Quantity: 10 distinct ways/methods.
- Depth: Extensive detail.
- Requirements: Include charts and tables.
- Goal: Reduce global temperatures (geoengineering/cooling technologies).
-
Deconstruct the Science:
- Concept: Radiative cooling is the process where an object loses heat by emitting thermal radiation (infrared light) to the cold sink of outer space (approx. 3K).
- Mechanism: The "atmospheric window" (8–13 µm). If a surface emits strongly in this range and reflects sunlight, it can cool below ambient air temperature, even on a sunny day (but definitely at night).
- Challenge: Scaling this up to "reduce global temperatures" vs. just cooling buildings. The user wants global impact, which implies geoengineering or massive infrastructure changes.
-
Brainstorming 10 Ways (Categorization Strategy):
- Urban/Building Scale (Passive): Roof coatings, windows, asphalt.
- Water/Agriculture Scale: Evaporative cooling, lake covers, crop selection.
- Geoengineering/Atmospheric Scale: High-altitude reflectors, modified cirrus clouds, stratospheric aerosols (though this is different from radiative cooling, it relates to albedo).
- Industrial/Power Scale: Cooling towers, waste heat dissipation.
Drafting the List:
- Urban Roof Coatings (High Albedo/IR emissivity).
- Smart Windows (Spectral selectivity).
- Radiative Cooling Asphalt (Roads).
- Nighttime Sky Radiators (Large scale ground-based emitters).
- Floating Radiative Coolers (Oceans).
- Agricultural Blankets (Crop protection).
- Power Plant Integration (Zero-water cooling towers).
- Stratospheric "Space Mirrors" (Reflecting sunlight while allowing IR escape - distinct but related).
- Cirrus Cloud Thinning (Allowing more IR to escape - distinct but related).
- Personal Radiative Cooling (Textiles).
Critique: Some of these (8 & 9) are more about solar radiation management (SRM) than purely "nighttime radiative cooling," but they manipulate the same energy balance. I should focus heavily on enhancing the escape of heat. Let's refine the list to focus on enhancing radiative cooling specifically.
Refined List:
- Super-Radiative Urban Roofing (Building-scale).
- Infra-Red Emissive Road Surfaces (Transport scale).
- Floating Ocean Radiators (Ocean-scale).
- Radiative Cooling Shading Structures (Infrastructure).
- Passive Nighttime Air Conditioning Systems (HVAC scale).
- High-Emissivity Agricultural Mulching (Agri-scale).
- Industrial Waste Heat Radiators (Manufacturing scale).
- Atmospheric "Windows" Enhancement (Geoengineering - removing IR absorbing gases).
- Space-Based Radiative Coolers (Orbital scale).
- Personal Radiative Cooling Fabrics (Human scale).
Correction on Global Impact: To truly address "reduce global temperatures," I need to acknowledge that building-scale cooling helps energy demand (reducing indirect emissions) but doesn't directly cool the planet unless the scale is massive (e.g., covering oceans). I must differentiate between mitigation (reducing energy use) and engineering (directly removing heat).
-
Developing Content (Extensive Detail):
- Mechanism explanation: Must define the 8-13 micrometer window.
- Physics: Stefan-Boltzmann law, emissivity vs. absorptivity.
Drafting specific points:
- Point 1 (Roofs): Focus on "Radiative Cooling Paint." Not just white paint, but IR-optimized paint.
- Point 2 (Roads): Dark asphalt is a heat trap. Replace with IR-emissive materials.
- Point 3 (Ocean): Water naturally cools at night. Can we enhance it? Floating panels with high IR emissivity.
- Point 4 (HVAC): Thermal Energy Storage (TES). Cool water at night using radiation, store ice/coolth, use during day.
- Point 5 (Ag): Frost protection is known. Can we use it to lower field temps?
- Point 6 (Industry): Use radiative cooling instead of water for condenser towers.
- Point 7 (Personal): Textiles that let body heat escape to space.
- Point 8 (Cirrus): While technically cloud thinning, it facilitates radiative cooling. (Keep this, it's a strong geoengineering link).
- Point 9 (Space Mirrors): Wait, mirrors reflect sun. How do they cool? By not blocking IR? Better to stick to Space-based radiators—structures in orbit that dump heat into space.
- Point 10 (The "Atmospheric Window" Shield): A conceptual geoengineering idea where we create a shield that blocks UV but lets IR through? No, that's confusing. Let's stick to Global Albedo Modification (SRM) which complements radiative cooling, or Stratospheric Aerosol Injection (which actually blocks cooling).
- Better Point 10: Nighttime Sky Radiators for Grid Stability. Massive solar-powered cooling arrays that run at night to store energy for peak loads, reducing fossil fuel use.
Let's pivot to the most scientifically rigorous interpretation of "using nighttime cooling": The core concept is Passive Radiative Cooling (PRC).
Revised List for maximum scientific depth:
- Radiative Cooling Paints for Urban Canopy: Enhancing emissivity of buildings.
- Infra-Red Transparent/Emisive Asphalt: Changing road thermodynamics.
- Ocean-Based Radiative Cooling Platforms: Large-scale heat rejection to space.
- Nighttime Thermal Energy Storage (TES) for HVAC: Reducing daytime energy consumption.
- Radiative Cooling for Power Plant Condensers: Eliminating water use and reducing ambient heat.
- High-Emissivity Agricultural Mulching: Managing microclimates.
- Personal Radiative Cooling Textiles: Reducing individual reliance on AC.
- Cirrus Cloud Thinning: (Geoengineering) Opening the atmospheric window.
- Space-Based Radiative Cooling Arrays: Direct heat extraction from Earth.
- Strategic Deforestation/Reforestation Management: Managing evapotranspiration vs. radiative cooling (complex interaction). Actually, let's swap this for something more direct. Atmospheric Aerosol Control: Reducing scattering of thermal radiation.
Self-Correction on Geoengineering: The prompt asks how to use the nighttime cooling to reduce temps. Most methods are local. I must explain the aggregate effect. If everyone uses cool roofs, the "Urban Heat Island" shrinks. That lowers regional temps.
-
Designing Visuals (Charts & Tables):
- Table 1: Comparison of methods (Scale, Cost, Efficiency, Global Potential).
- Chart 1: Conceptual diagram of the Radiative Cooling Mechanism (The 8-13 µm window).
- Chart 2: Impact estimation (Hypothetical temperature reduction curves).
-
Writing the Content (Iterative refinement):
- Introduction: Define the physics clearly. The "Atmospheric Window."
- Body Paragraphs: Use technical terms (Planck's law, emissivity $\epsilon$, solar absorptivity $\alpha_{solar}$).
- Detailing each point:
- Roofs: Mention recent research (e.g., Berkeley Lab). $\Delta T$ of 5-10°C.
- Ocean: Mention the heat capacity issue.
- Power Plants: Mention Rankine cycle efficiency.
- Cirrus Cloud Thinning: This is crucial. Cirrus clouds trap heat. Removing them allows the natural radiative cooling to work faster. This is a direct link to the prompt.
-
Review against Constraints:
- 10 ways? Yes.
- Extensive detail? Yes, need to ensure physics is explained.
- Charts/Tables? Yes, need to format these clearly using Markdown.
- Global focus? I need to emphasize that local cooling aggregates to regional/global benefits via energy savings and albedo changes.
-
Final Polish:
- Ensure the distinction between "Passive Radiative Cooling" (material science) and "Solar Radiation Management" (geoengineering) is clear, but show how they utilize the same thermodynamic principles.
- Check tone: Informative, scientific, objective.
(Self-Correction during drafting Point 8): Cirrus Cloud Thinning is technically removing an obstacle to radiative cooling. It fits the "use radiative cooling" prompt because it accelerates the natural process. (Self-Correction during drafting Point 9): Space-based radiators are theoretical but fit the "extensive detail" requirement as a high-concept solution.
-
Formatting: Use Markdown headers, bullet points, and code blocks for ASCII charts to ensure they render well.
(Plan for ASCII Charts):
- One showing the spectral emissivity curve (solar reflection vs IR emission).
- One showing the comparison matrix.
-
Final Output Generation. (Proceeding to generate response based on these steps.)
</think>
Here is an extensive breakdown of 10 distinct methods for utilizing nighttime radiative cooling—specifically the process of emitting thermal infrared radiation into the deep cold of space—to mitigate global temperatures.
The Physics of Radiative Cooling
Before exploring the methods, it is essential to understand the mechanism. Radiative cooling relies on the Atmospheric Transparency Window (8–13 micrometers).
- The Concept: The Earth’s atmosphere is opaque to most infrared (IR) radiation due to greenhouse gases ($CO_2$, water vapor), trapping heat. However, between 8 and 13 µm, the atmosphere is largely transparent.
- The Sink: Outer space acts as a thermal sink at approximately 3 Kelvin (-270°C).
- The Process: If a surface is engineered to have high emissivity ($\epsilon$) in the 8–13 µm range and high reflectivity (albedo) in the solar spectrum, it can emit thermal energy directly into space, bypassing the greenhouse effect. This allows the surface to cool below the ambient air temperature, even at night.
1. Global Deployment of Super-Radiative Roofing
The most immediate application is replacing standard roofing materials with Passive Radiative Cooling (PRC) materials.
- Mechanism: Traditional white paint reflects sunlight (albedo ~0.9) but absorbs heat in the infrared spectrum. Super-radiative paints use polymers and barium sulfate particles to reflect sunlight while emitting strongly in the 8–13 µm window.
- Implementation: Global retrofitting of commercial and residential buildings.
- Impact on Global Temps:
- Direct: Reduces the Urban Heat Island (UHI) effect. Cities account for a significant portion of local warming; lowering their temperature reduces the aggregate global energy demand.
- Indirect: Reduces electricity consumption for air conditioning by 20–40%, thereby lowering $CO_2$ emissions from power plants, which enhances the atmospheric window further.
- Technical Specs: Achieves sub-ambient cooling of $\Delta T \approx 5-8°C$.
2. Infra-Red Emissive Road Surfaces
Asphalt is a primary driver of urban heating due to its low albedo and low IR emissivity.
- Mechanism: Developing "Cool Pavement" composites that incorporate phase-change materials (PCMs) and IR-emissive ceramics.
- Implementation: Laying these materials in major highway networks and urban centers globally.
- Impact:
- Nighttime Sink: During the night, these roads actively radiate the heat absorbed during the day directly into space, rather than radiating it back into the street-level air.
- Albedo: Increases the planetary albedo of urban infrastructure, reflecting more solar energy before it is absorbed.
3. Floating Ocean Radiative Cooling Platforms
The oceans absorb the vast majority of excess heat in the climate system. Enhancing their ability to shed this heat is critical.
- Mechanism: Deploying large-scale floating arrays (similar to solar farms) covered in high-emissivity hydrophobic membranes.
- Implementation: Placing these platforms in warm ocean currents (e.g., the Gulf Stream, Kuroshio).
- Process: The platform floats on the water. At night, the membrane radiates heat into space. Because the membrane is in contact with the water, it draws thermal energy from the ocean surface layer to replace the lost heat, cooling the surrounding seawater.
- Impact: Reduces the thermal load of the oceans, potentially slowing the release of heat back into the atmosphere and reducing ocean acidification rates.
4. Nighttime Thermal Energy Storage (TES) for HVAC
Instead of just cooling buildings, we can use radiative cooling to store energy, reducing the need for fossil-fuel-based power generation.
- Mechanism: Large-scale radiative cooling towers are installed. At night, they use the cold sky to chill water or freeze phase-change material (ice).
- Implementation: Replacing traditional water-intensive cooling towers with dry radiative cooling towers in industrial complexes.
- Impact:
- Peak Shaving: The stored "coolth" is used during the day to air-condition cities. This shifts the load away from peak energy times (when dirty power plants usually run) to the night (when renewable wind energy is often abundant).
5. Radiative Cooling for Power Plant Condensers
Power plants (nuclear, coal, natural gas) rely on cooling towers to condense steam, often releasing massive plumes of hot water and heat into the environment.
- Mechanism: Replace water-based condenser cooling with radiative cooling condensers.
- Process: These systems use a vacuum-insulated panel with a high-emissivity surface facing the night sky. Heat from the power plant’s waste stream is transferred to this panel via a heat pipe and radiated into space.
- Impact:
- Zero-Water Usage: Solves the conflict between energy production and water scarcity.
- Global Temp: Reduces the thermal pollution of local water bodies and the atmosphere, maintaining the efficiency of the Rankine cycle while preventing excess waste heat accumulation.
6. Cirrus Cloud Thinning (Geoengineering)
While this is a geoengineering technique, it is fundamentally about enhancing radiative cooling.
- The Problem: High-altitude Cirrus clouds act like a blanket. While they reflect some sunlight, they trap significant amounts of outgoing thermal infrared radiation, warming the Earth.
- The Method: Using aircraft to inject ice nuclei (like silver iodide) into supercooled cirrus clouds. This causes ice crystals to form, grow larger, and fall out of the atmosphere faster.
- Result: Thinner cirrus clouds allow more terrestrial infrared radiation (heat) to escape through the atmospheric window into space.
- Impact: This is one of the most effective ways to utilize the "nighttime cooling" mechanism on a planetary scale, directly reversing the radiative forcing caused by aviation and natural cloud thickening.
7. High-Emissivity Agricultural Mulching
Agriculture accounts for a large land area. Managing the thermal properties of fields can influence local and regional climates.
- Mechanism: Using plastic or biodegradable mulch films with specific optical properties.
- Design: The film must reflect visible light (to keep crops cool) but emit strongly in the 8–13 µm range.
- Process: At night, the film radiates heat away from the soil and crop canopy into space.
- Impact:
- Frost Protection: Historically used to stop frost, but optimized for cooling, it can reduce heat stress on crops during heatwaves.
- Reduced Irrigation: Lower soil temperatures reduce evaporation rates, conserving water and reducing the latent heat flux into the atmosphere.
8. Personal Radiative Cooling Textiles
Global temperature reduction is partly dependent on reducing the total energy demand of the human population.
- Mechanism: Weaving fabrics with hollow fiber structures or polymer coatings that have high transmittance in the mid-infrared (MIR) range.
- Process: When a person wears this clothing, their body heat (emitted at ~9-10 µm) passes through the fabric and radiates directly into the sky (or the immediate environment) rather than being trapped by the fabric layers.
- Impact:
- Comfort: Allows people to remain comfortable at higher ambient temperatures without using air conditioning.
- Systemic: If adopted globally, this reduces the "thermal load" on urban environments, as less heat is generated by HVAC exhaust fans.
9. Space-Based Radiative Cooling Arrays
A theoretical but scientifically grounded method for large-scale heat rejection.
- Concept: Deploying massive radiators in Low Earth Orbit (LEO) or Geostationary Orbit (GEO).
- Mechanism: These structures would be connected to Earth’s energy grid (via wireless power transfer or simply intercepting solar energy). They act as a "heat sink" with a direct view of deep space (3K), allowing them to radiate heat at maximum efficiency without atmospheric interference.
- Application: They could be used to dump waste heat from solar power satellites or, in a geoengineering context, to actively cool the upper atmosphere.
10. Atmospheric Aerosol Control (Cleaning the Window)
Radiative cooling is stifled by pollution. To "use" nighttime cooling effectively, we must clear the path.
- The Problem: Sulfate aerosols and black carbon (soot) scatter and absorb infrared radiation, blocking the atmospheric window.
- The Method: Aggressive reduction of short-lived climate pollutants (SLCPs), specifically methane and black carbon, and sulfate emissions.
- Mechanism: By cleaning the atmosphere, we increase its transparency in the 8–13 µm window.
- Impact: This allows the Earth’s natural radiative cooling processes to function at peak efficiency, accelerating the dissipation of heat at night without altering solar input.
Comparative Analysis of Methods
The following table compares the feasibility, scale, and cooling potential of these methods.
| Method | Scale of Impact | Tech Readiness (TRL) | Primary Mechanism | Cost |
|---|---|---|---|---|
| 1. Roof Paints | High (Urban) | High (TRL 8-9) | High Albedo + High IR Emissivity | Low |
| 2. Cool Asphalt | Medium (Urban) | Medium (TRL 6-7) | Phase Change + IR Emission | Medium |
| 3. Ocean Floaters | Global | Low (TRL 3-4) | Direct Ocean Heat Extraction | High |
| 4. TES HVAC | Regional | Medium (TRL 7) | Energy Storage (Coolth) | Medium |
| 5. Power Plant Radiators | Regional | Medium (TRL 6) | Zero-Water Heat Dissipation | High |
| 6. Cirrus Thinning | Global | Low (TRL 3-4) | Enhancing IR Escape (Geoeng) | Medium |
| 7. Ag Mulching | Regional | High (TRL 9) | Microclimate Management | Low |
| 8. Cool Textiles | Individual | Medium (TRL 6-7) | Personal Heat Rejection | Low |
| 9. Space Radiators | Global | Very Low (TRL 1-2) | Direct Space Heat Dump | Very High |
| 10. Aerosol Control | Global | High (TRL 9) | Increasing Atmospheric Trans. | N/A (Policy) |
Charts: Visualizing the Science
Chart 1: The Radiative Cooling Mechanism
This chart illustrates the difference between a standard surface and a radiative cooling surface.
RADIATIVE SPECTRUM (Wavelength in Micrometers)
SOLAR INPUT THERMAL EMISSION
(0.3 - 2.5 µm) (3 - 25 µm)
|-----------------| |-----------------------------|
0 2.5 5 15
STANDARD SURFACE:
(e.g., Dark Asphalt)
[ ABSORB ] [ ABSORB ]
(High Solar) (Medium IR)
Result: Heats up rapidly, radiates heat into atmosphere.
RADIATIVE COOLING SURFACE:
(e.g., Super-Radiative Coating)
[ REFLECT ] [ EMIT STRONGLY ]
(High Albedo) (High Emissivity in 8-13 µm window)
^ ^
| |
| |--> Heat escapes to SPACE (3K)
Result: Stays cool, actively loses heat to space.
Chart 2: Estimated Temperature Reduction Potential
A comparative visualization of how much a specific method can lower temperature relative to the baseline ambient temperature.
| Method | Potential Sub-Ambient Cooling ($\Delta T$) | Primary Application Area |
|---|---|---|
| Super-Radiative Roof | -5°C to -8°C | Residential/Commercial |
| Road Surfaces | -3°C to -5°C | Infrastructure |
| Power Plant Radiators | -10°C (Waste Heat) | Industrial |
| Cool Textiles | -3°C (Skin Temp) | Personal |
| Cirrus Thinning | +0.5°C Global Cooling (Est.) | Planetary |
Note: "Sub-ambient" means the surface temperature is lower than the surrounding air.
Conclusion
While methods 1 through 5 are localized, their global adoption creates a "cooling cascade" by reducing the energy required to fight global warming. Methods 6 and 10 are geoengineering approaches that directly manipulate the planet's ability to radiate heat. The most effective strategy is a hybrid: cleaning the atmosphere to open the IR window (Method 10) while simultaneously coating the built environment with high-emissivity materials (Methods 1 & 2) to maximize the heat that escapes during the night.
EXAMPLE #2
User
Come up with six plots for a new "Black Mirror" episode (that the audience would love) that all involve time travel with sexy theme(s). Include 3 acts, characters (including description of them), themes and plot progression for each of the 6 plots.
Assistant
<think>
Here's a thinking process that leads to the suggested plots:
-
Deconstruct the Request:
- Topic: Six plots for a new Black Mirror episode.
- Core Element: Time travel.
- Core Element: "Sexy" theme(s) (romance, desire, attraction, sexuality, relationships).
- Requirement: Audience would love them (needs to fit the show's DNA: dystopian, psychological, twist-heavy, thought-provoking, emotional).
- Structure: 3 Acts, Characters (with descriptions), Themes, Plot Progression for each of the 6.
-
Analyze "Black Mirror" Style:
- Tone: Cold, cynical, often tragic, occasionally hopeful but usually at a high cost. High-tech, low-life.
The model card is truncated. Read the rest on Hugging Face.
- Task
- Image-Text-to-Text
- Type
- qwen3_5
- License
- apache-2.0
- Languages
- en, zh
- Created
- Jul 28, 2026
- Updated
- Jul 28, 2026