Ornith-1.5-35B-A3B-NVFP4-MTP

jedi-95/Ornith-1.5-35B-A3B-NVFP4-MTP

Ornith-1.5-35B-A3B-NVFP4-MTP is a text generation model from jedi-95. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as mit.

<img width="600px" src="assets/ornithlogo.png"

Downloads · 30 days

106

11% of all-time downloads

All-time downloads

934

Public

Parameters

18.7B

23.4 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors23.4 GB · 100%

Parameter types

How the weights are stored.

U816.4B · 84%

Try a prompt

Base models

Model card

From the Hugging Face model README

Ornith-1.5-35B-A3B-NVFP4 (Updated MTP)

Community MTP-weight update for the official NVFP4 checkpoint. This is not an official Ornith release. It is a surgical weight swap that replaces the MTP (Multi-Token Prediction) head in the official ornith-ai/Ornith-1.5-35B-A3B-NVFP4 checkpoint with the updated MTP weights from the BF16 base model (ornith-ai/Ornith-1.5-35B-A3B, shard 16 re-uploaded 2026-08-22).

Why this exists

On 2026-08-24 Ornith announced updated MTP weights for all Ornith-1.5 variants (BF16, GGUF, NVFP4, FP8). The BF16 and FP8 repos received fresh weight commits, but the NVFP4 repo did not — its last weight commit predates the MTP update. A user asked about this on the NVFP4 repo; as of this upload no official update has been pushed.

This repo fills that gap.

What changed

Only the MTP head was replaced. Everything else is byte-identical to the official NVFP4 checkpoint.

The MTP head consists of 785 tensors (~1.69 GB): a single MoE layer with 256 routed experts, a shared expert, self-attention, and the FC/norm projections. Critically, the official NVFP4 quant config excludes MTP from quantization (exclude_modules: ["mtp*", "mtp.layers.0*"]), so these tensors are stored as unquantized BF16 in both the NVFP4 and BF16 checkpoints. The swap is therefore a direct BF16→BF16 tensor replacement — no requantization, no dequant/requant, no format conversion.

	Official NVFP4	This repo
Non-MTP weights	NVFP4 / FP8 (quantized)	Identical (byte-for-byte)
MTP weights	BF16 (old, pre-update)	BF16 (updated, from BF16 base shard 16)
Config / tokenizer / quant config	Original	Identical
Total size	23.41 GB	23.42 GB

Verification

The swap was verified at the binary level:

7,774 tensors in shard 3: same names, shapes, and dtypes as the original.
6,989 non-MTP tensors: byte-identical to the original NVFP4 shard (verified by reading and comparing every tensor's raw bytes).
785 MTP tensors: all 785 changed (new weights from the BF16 base model).
MTP dtype: all BF16 (unchanged — MTP was never quantized).
Non-MTP dtypes: BF16, F8_E4M3, U8, F32 (unchanged — quantized weights untouched).

How it was done

Downloaded the official NVFP4 repo and BF16 base model shard 16.
Extracted the 785 mtp.* tensors from BF16 shard 16.
Replaced the 785 mtp.* tensors in NVFP4 shard 3, preserving all 6,989 non-MTP tensors byte-for-byte.
Rebuilt the safetensors file with corrected data_offsets in the header.
Updated model.safetensors.index.json with the correct total_size.

The swap script is available at swap_mtp.py (reproduced in the Files tab).

Serving

Serve this exactly as you would the official NVFP4 checkpoint. The only difference is the MTP weights — all configs, tokenizer, chat template, and quant configs are unchanged.

vllm serve jedi-95/Ornith-1.5-35B-A3B-NVFP4-MTP     --served-model-name Ornith-1.5-35B-A3B     --tensor-parallel-size 1     --max-model-len 262144     --gpu-memory-utilization 0.90     --spec-method mtp --spec-tokens 1     --enable-prefix-caching     --enable-auto-tool-choice --tool-call-parser qwen3_xml     --reasoning-parser qwen3     --trust-remote-code

For DGX Spark (GB10) specifically, see MiaAI-Lab/Ornith-1.5-35B-A3B-DGX-Spark for the b12x backend recipe and CUDA-graph patches.

Disclaimer

This is an unofficial community contribution. The model architecture, quantization, benchmarks, and all non-MTP weights are the work of the Ornith team. All credit for the model belongs to them. This repo only swaps the MTP head to match the updated BF16/FP8 checkpoints. If Ornith pushes an official NVFP4 update, this repo becomes obsolete — use the official one instead.

License: MIT (same as the original).

Ornith-1.5-35B-A3B

Chirp Chirp! 🐦 We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.

Ornith-1.5 extends Ornith-1.0 (which was developed on top of Qwen3.5 and Gemma4 with additional continued pretraining, mid-training, and post-training) by expanding the self-improvement loop from scaffold and rollout optimization to jointly optimizing task generation, scaffold construction, and solution rollouts. Rather than relying on a fixed set of human-curated tasks and manually designed harnesses, Ornith-1.5 continuously generates new training tasks, discovers effective strategies for solving them, and improves the policy through reinforcement learning. For more details on the task, harness, and rollout reward design, please refer to our blog. <img style="width: 100%; max-width: 900px;" src="assets/ornith_35b_eval.png" alt="Ornith 1.5 35B Benchmark Results" title="Ornith 1.5 35B Benchmark Results">

Ornith 1.5 35B-A3B

This model card documents Ornith-1.5-35B-A3B, the mid-size mixture-of-experts member of the Ornith-1.5 family. It activates only ~3B parameters per token, yet significantly outperforms its similar-sized peer Qwen 3.6-35B across all coding and agentic benchmarks, and outperforms dense models such as Gemma 4-31B and Muse Glimmer-30B by wide margins on agentic coding.

Benchmarks

<div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;width:100%;margin:0 auto;padding:16px 0"> <table style="width:100%;table-layout:fixed;border-collapse:collapse;font-size:13px"> <thead><tr> <th style="width:22%;padding:10px 7px;text-align:left;font-weight:600;border-bottom:2px solid #FD8E5B;color:#FD8E5B"></th> <th style="width:13%;padding:10px 7px;text-align:center;font-weight:700;border-bottom:2px solid #FD8E5B;color:#FD8E5B;font-size:14px;background:rgba(253, 142, 91, 0.12)">Ornith-1.5-35B-A3B</th> <th style="width:13%;padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #FD8E5B;color:#FD8E5B;font-size:14px">Ornith-1.0-35B-A3B</th> <th style="width:13%;padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #FD8E5B;color:#FD8E5B;font-size:14px">Qwen3.6-35B-A3B</th> <th style="width:13%;padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #FD8E5B;color:#FD8E5B;font-size:14px">Gemma-4-31B <sub></sub></th> <th style="width:13%;padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #FD8E5B;color:#FD8E5B;font-size:14px">Muse-Glimmer-30B <sub></sub></th> <th style="width:13%;padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #FD8E5B;color:#FD8E5B;font-size:14px">Qwen3.5-397B</th> </tr></thead> <tbody> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#FD8E5B;border-bottom:1px solid rgba(253, 142, 91, 0.2);background:rgba(253, 142, 91, 0.1)">Coding</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Terminal-Bench 2.1 <sub><small>(Terminus-2)</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">67.8</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)">52.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">42.1</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)">53.5</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Terminal-Bench 2.1 <sub><small>(Claude Code)</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">68.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">62.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">49.2</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)">48.6</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);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">79</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.6</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)">52</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.4</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);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">59.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.4</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)">35.7</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)">51.6</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);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">71.4</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)">67.2</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)">-</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.3</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">DeepSWE</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">22</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">0</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">0</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)">1</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Frontier-Bench v0.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">5.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1.4</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)">-</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1.4</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);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">46.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">34.6</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)">15.5</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)">36.8</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE Atlas - QnA</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">39.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">37.1</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)">-</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)">20.4</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#FD8E5B;border-bottom:1px solid rgba(253, 142, 91, 0.2);background:rgba(253, 142, 91, 0.1)">Reasoning</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HLE <sub><small>(no tools)</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">25.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">20.8</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)">19.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">22</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HLE <sub><small>(with tools)</small></sub></td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">33.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">30.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.9</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">26.5</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)">48.3</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);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">89.2</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</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)">83.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.4</td> </tr> <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#FD8E5B;border-bottom:1px solid rgba(253, 142, 91, 0.2);background:rgba(253, 142, 91, 0.1)">Agentic</td></tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MCP-Atlas</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">70.2</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">62.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">55</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.3</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Toolathlon-Verified</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">48.7</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">42.4</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)">40.8</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)">38.3</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">WideSearch</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">67.8</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">63.4</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.1</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">54.2</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)">74</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">BrowseComp</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">67.6</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">63.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">62</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)">78.6</td> </tr> <tr> <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">ClawEval</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15);font-weight:600;color:#FD8E5B;background:rgba(253, 142, 91, 0.06)">72.5</td> <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.8</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)">48.5</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)">70.7</td> </tr> </tbody> </table> <p style="margin-top:12px;font-size:10px;opacity:0.7"> * All results reported for Ornith-1.5 are averaged over five independent runs.<br/> * Terminal-Bench 2.1 (Terminus-2): We evaluate Terminal-Bench 2.1 using the Harbor/Terminus-2 framework with parser=json, temperature=1.0, top_p=1.0, and a 128K context window. Each run uses a 4-hour timeout with 32 CPU cores and 48GB RAM, and results are averaged over 5 runs. We adjust the Qwen chat template to ensure consistency between training and inference (https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/chat_template.jinja), and modify Harbor to align with vLLM's reasoning_content key.<br/> * Terminal-Bench 2.1 (Claude Code): We evaluate Terminal-Bench 2.1 using Claude Code 2.1.126 with parser=json, temperature=1.0, top_p=1.0, max_new_tokens=131072. Results are averaged over 5 runs. Again, Qwen chat template needs to be modified.<br/> * SWE-Bench Verified, Pro and Multilingual: using OpenHands harness with temp=1.0, top_p=0.95, 256k context window. Anti-hacking safeguards are applied throughout evaluation: Git history is removed from the local repository image to prevent access to prior solutions or commits; network access is disabled, preventing the model from retrieving external information or resources.<br/> * DeepSWE: Evaluated using the Claude Code harness with temperature=1.0, top_p=0.95, and a 256K context window.<br/> * SWE Atlas QnA: using mini SWE agent harness with temp=1.0, top_p=0.95, 128K context window. Results are averaged over 5 runs.<br/> * NL2Repo: with temperature=1.0, top_p=1.0, 400K context, 48K output. Access to specified GitHub repositories and pip packages is blocked to prevent reward hacking.<br/> * HLE: Evaluated using Claude 4.6 Opus as the judge model.<br/> * MCP-Atlas: All models were evaluated in thinking mode on the 500-task public subset, with a 10-minute timeout per task. We use Claude 4.8 Opus as the judge model.<br/> * Toolathlon-Verified: We use the official evaluation service with the maximum token limit set to 128K.<br/> * ClawEval: An agentic code benchmark over real-user task distributions; temp=0.6 and 256K context.<br/> </p> </div>

Quickstart

<div style="border-left:4px solid #FD8E5B;background:rgba(253,142,91,0.1);border-radius:6px;padding:12px 16px;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;font-size:14px;line-height:1.6"> <div style="font-weight:700;color:#FD8E5B;margin-bottom:6px">📝 NOTE</div> <p style="margin:0 0 10px"><b>Ornith-1.5-35B-A3B</b> is a <b>reasoning model</b>: by default the assistant turn opens with a <code style="background:rgba(253,142,91,0.15);padding:1px 5px;border-radius:4px"><think> … </think></code> block before the final answer. The serving recipes below enable a reasoning parser so the chain-of-thought is returned in a separate <code style="background:rgba(253,142,91,0.15);padding:1px 5px;border-radius:4px">reasoning_content</code> field, and a tool-call parser so the model's <code style="background:rgba(253,142,91,0.15);padding:1px 5px;border-radius:4px"><tool_call></code> blocks are surfaced as OpenAI-style <code style="background:rgba(253,142,91,0.15);padding:1px 5px;border-radius:4px">tool_calls</code>.</p> <p style="margin:0 0 6px">Serving Ornith-1.5-35B-A3B requires recent runtimes:</p> <ul style="margin:0 0 10px;padding-left:20px"> <li><b>Transformers</b> ≥ 5.8.1</li> <li><b>vLLM</b> ≥ 0.19.1</li> <li><b>SGLang</b> ≥ 0.5.9</li> </ul> <p style="margin:0 0 6px">Recommended sampling parameters:</p> <ul style="margin:0;padding-left:20px"> <li><b>For general tasks:</b> <code style="background:rgba(253,142,91,0.15);padding:1px 5px;border-radius:4px">temperature=0.6</code>, <code style="background:rgba(253,142,91,0.15);padding:1px 5px;border-radius:4px">top_p=0.95</code>, <code style="background:rgba(253,142,91,0.15);padding:1px 5px;border-radius:4px">top_k=20</code></li> <li><b>To reproduce the reported benchmarks:</b> <code style="background:rgba(253,142,91,0.15);padding:1px 5px;border-radius:4px">temperature=1.0</code></li> </ul> </div>

Serving Ornith-1.5-35B-A3B

Ornith-1.5-35B-A3B is a ~35B mixture-of-experts model with ~3B activated parameters per token (≈70 GB in bf16). The recipes below stand up an OpenAI-compatible server on 2× 80GB GPUs to leave headroom for the 256K context; adjust --tensor-parallel-size / --tp to match your hardware.

vLLM

vllm serve ornith-ai/Ornith-1.5-35B-A3B \
    --served-model-name Ornith-1.5-35B-A3B \
    --host 0.0.0.0 --port 8000 \
    --tensor-parallel-size 2 \
    --max-model-len 262144 \
    --gpu-memory-utilization 0.90 \
    --enable-prefix-caching \
    --enable-auto-tool-choice --tool-call-parser qwen3_xml \
    --reasoning-parser qwen3 \
    --trust-remote-code

SGLang

python -m sglang.launch_server \
    --model-path ornith-ai/Ornith-1.5-35B-A3B \
    --served-model-name Ornith-1.5-35B-A3B \
    --host 0.0.0.0 --port 8000 \
    --tp 2 \
    --context-length 262144 \
    --mem-fraction-static 0.85 \
    --tool-call-parser qwen3_coder \
    --reasoning-parser qwen3

For Long-Context

Ornith-1.5-35B-A3B handles context windows of up to 262,144 tokens. When a task's combined input and output must go beyond this limit, we suggest extending the effective window with RoPE scaling — YaRN is the technique we validate against, and it is already built into both vLLM and SGLang. With a scaling factor of 4.0, the usable window grows to roughly 1M tokens.

You can turn YaRN on in either of two ways:

Edit the checkpoint's config.json. Add a rope_scaling block to the model configuration:

{
    "rope_scaling": {
        "rope_type": "yarn",
        "factor": 4.0,
        "original_max_position_embeddings": 262144
    }
}

Override at launch time. Leave the checkpoint untouched and extend the serve commands above with the equivalent flags.

vLLM:

VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ornith-ai/Ornith-1.5-35B-A3B ... --hf-overrides '{"rope_scaling": {"rope_type": "yarn", "factor": 4.0, "original_max_position_embeddings": 262144}}' --max-model-len 1000000

SGLang:

SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"rope_scaling": {"rope_type": "yarn", "factor": 4.0, "original_max_position_embeddings": 262144}}' --context-length 1000000

Using Ornith-1.5-35B-A3B via the Chat Completions API

Once a vLLM or SGLang server is running, talk to it with any OpenAI-compatible client.

Basic Usage

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="EMPTY",  # any non-empty string works for a local server
)

response = client.chat.completions.create(
    model="Ornith-1.5-35B-A3B",
    messages=[
        {"role": "user", "content": "Write a one-line Python lambda that squares a number."}
    ],
    temperature=0.6,
    top_p=0.95,
    max_tokens=1024,
)

message = response.choices[0].message
# reasoning_content holds the <think> trace; content holds the final answer.
print("reasoning:", getattr(message, "reasoning_content", None))
print("answer:", message.content)

You can also stream tokens, or hand the model tools — Ornith-1.5-35B-A3B emits well-formed function calls that the server parses into the standard tool_calls field:

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get the current weather for a city",
            "parameters": {
                "type": "object",
                "properties": {"city": {"type": "string"}},
                "required": ["city"],
            },
        },
    }
]

response = client.chat.completions.create(
    model="Ornith-1.5-35B-A3B",
    messages=[{"role": "user", "content": "What is the weather in Paris right now?"}],
    tools=tools,
    tool_choice="auto",
    temperature=0.6,
    max_tokens=2048,
)

tool_call = response.choices[0].message.tool_calls[0]
print(tool_call.function.name, tool_call.function.arguments)
# -> get_weather {"city": "Paris"}

You can point any OpenAI-compatible SDK (Python, Node.js, etc.) or curl at the same /v1/chat/completions endpoint.

Agentic Usage

Ornith-1.5-35B-A3B excels in tool-calling and agentic coding. It exposes an OpenAI-compatible endpoint with tool calling and works out of the box with standard agent frameworks.

Examples of using Ornith with agents:

Ollama

ollama run hf.co/ornith-ai/Ornith-1.5-35B-A3B-GGUF

Atomic.chat

# Atomic.chat loads a GGUF build of Ornith (ornith-ai/Ornith-1.5-35B-A3B-GGUF)
# through llama.cpp's OpenAI-compatible API on port 8000.
llama-server -hf ornith-ai/Ornith-1.5-35B-A3B-GGUF --port 8000 -c 262144

llama.cpp

# llama.cpp — serve an OpenAI-compatible API on port 8000.
llama-server -hf ornith-ai/Ornith-1.5-35B-A3B-GGUF --port 8000 -c 262144

Hermes Agent

# Hermes talks to any OpenAI-compatible endpoint — point it at your Ornith server.
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"
export MODEL="ornith-ai/Ornith-1.5-35B-A3B"

OpenClaw

# OpenClaw talks to any OpenAI-compatible endpoint — point it at your Ornith server.
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"
export OPENAI_MODEL="ornith-ai/Ornith-1.5-35B-A3B"

Unsloth Studio

pip install unsloth

# Load Ornith for fast local inference or fine-tuning (Python):
#   from unsloth import FastLanguageModel
#   model, tokenizer = FastLanguageModel.from_pretrained(
#       "ornith-ai/Ornith-1.5-35B-A3B",
#       max_seq_length=262144,
#       load_in_4bit=True,
#   )

Coding CLIs

Ornith-1.5-35B-A3B is optimized for terminal-based coding agents. Point any OpenAI-compatible coding CLI at your Ornith-1.5-35B-A3B endpoint (set OPENAI_BASE_URL and OPENAI_API_KEY) to understand large codebases, automate tedious work, and ship faster.

OpenCode

# Register your local Ornith endpoint as a provider in ~/.config/opencode/opencode.json:
#
# {
#   "$schema": "https://opencode.ai/config.json",
#   "provider": {
#     "ornith": {
#       "npm": "@ai-sdk/openai-compatible",
#       "name": "Ornith (local)",
#       "options": { "baseURL": "http://localhost:8000/v1", "apiKey": "EMPTY" },
#       "models": { "ornith-ai/Ornith-1.5-35B-A3B": { "name": "Ornith-1.5-35B-A3B" } }
#     }
#   }
# }

opencode

Citation

If you find our work helpful, feel free to give us a cite.

@misc{ornith_1_5,
    title = {{Ornith-1.5}: From Self-Scaffolding to Self-Improvement},
    url = {https://ornith.ai/ornith_1_5.html},
    author = {{Ornith Team}},
    year = {2026}
}

Task: Text Generation
Library: transformers
Type: qwen3_5_moe
License: mit
Created: Aug 26, 2026
Updated: Aug 26, 2026