Solar-Open2-Apex-GGUF

Solar Open 2

Solar Open 2 is Upstage’s 250B-A15B open-weight large language model, built for agentic use cases such as office productivity, document-intensive work, and coding. Its Hybrid-Attention Mixture-of-Experts (MoE) architecture with linear attention delivers highly efficient inference even in long-context settings.

Technical Report | Blog | Upstage Website | Try Demo (~7/31)

Highlights

• Agentic Specialist: Purpose-built for agentic workflows — tool calling, multi-step reasoning, and end-to-end task execution. Competitive with the strongest open-weight models on agent benchmarks.

• Minimal Inference Cost: A 250B-parameter MoE that activates only 15B per token, built on a hybrid attention stack that interleaves three linear-attention layers with one softmax-attention layer — large-model capacity at small-model inference cost.

• 1M-Token Context: The linear-attention layers encode token order intrinsically in their recurrent state, so positional encoding is removed entirely (NoPE), lifting the RoPE extrapolation limit. Only 12 of the 48 layers keep a KV cache, holding long-context memory to roughly a quarter of an all-softmax model of the same shape.

• Efficiently Trained at Low Cost: Initialized by selective weight transfer from Solar Open 1 (102B) — only the 2.3% of weights that survive the architectural change are carried over, and everything else is randomly initialized — which raises the starting point and accelerates early convergence at 250B scale.

• Multilingual: English, Korean, and Japanese.

---

Model Overview

---

Performance

English Benchmarks

Korean Benchmarks

† in-house benchmarks.

---

Quickstart

The examples below assume 8 GPUs with at least 141 GB of memory each, such as NVIDIA H200 or B200 GPUs. Actual memory requirements depend on the context length and serving settings.

Transformers

Use the Upstage Transformers branch with native Solar Open 2 support for local experimentation. For production serving, we recommend vLLM.

Install the dependencies:

Install a CUDA-enabled PyTorch build for your platform before running this command. fla-core enables the optimized KDA kernels; without it, Transformers uses a substantially slower PyTorch fallback.

python -m pip install -U \
  "git+https://github.com/upstageAI/[email protected]" \
  "fla-core[cuda]>=0.5.1" \
  accelerate einops

Run the model:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "upstage/Solar-Open2-250B"

tokenizer = AutoTokenizer.from_pretrained(
    model_id,
    trust_remote_code=False,
)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    dtype=torch.bfloat16,
    trust_remote_code=False,
)
model.eval()

messages = [
    {"role": "user", "content": "What is Upstage?"},
]
prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    reasoning_effort="high",
    think_render_option="preserved",
)
input_device = model.get_input_embeddings().weight.device
model_inputs = tokenizer(prompt, return_tensors="pt").to(input_device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768,
    do_sample=True,
    temperature=1.0,
    top_p=1.0,
)

new_token_ids = generated_ids[0, model_inputs.input_ids.shape[-1] :].tolist()
think_end_id = tokenizer.convert_tokens_to_ids("<|think:end|>")

if think_end_id in new_token_ids:
    # Split immediately after the final <|think:end|> token.
    answer_start = len(new_token_ids) - new_token_ids[::-1].index(think_end_id)
else:
    # No end marker usually means generation stopped while the model was reasoning.
    answer_start = len(new_token_ids)

reasoning = tokenizer.decode(
    new_token_ids[:answer_start],
    skip_special_tokens=True,
).strip()
answer = tokenizer.decode(
    new_token_ids[answer_start:],
    skip_special_tokens=True,
).strip()

print("[reasoning]", reasoning)
print("[answer]", answer)

If the answer is empty, generation likely reached max_new_tokens before the reasoning block ended. Increase max_new_tokens and try again.

Serving with vLLM (Recommended)

Option 1: Docker

The image below is based on vLLM v0.22.0 and CUDA 12.9.

docker run --rm --gpus all --ipc=host \
  -p 8000:8000 \
  -v "${HF_HOME:-$HOME/.cache/huggingface}:/root/.cache/huggingface" \
  upstage/vllm-solar-open2 \
  upstage/Solar-Open2-250B \
  --served-model-name solar-open2-250b \
  --tensor-parallel-size 8 \
  --enable-expert-parallel \
  --moe-backend triton \
  --default-chat-template-kwargs '{"think_render_option":"preserved"}' \
  --reasoning-parser solar_open2 \
  --tool-call-parser solar_open2 \
  --enable-auto-tool-choice \
  --logits-processors vllm.v1.sample.logits_processor.solar_open2:SolarOpen2TemplateLogitsProcessor

Option 2: Install from source

Install the Upstage fork while reusing the matching vLLM v0.22.0 CUDA 12.9 wheel:

pip install -U uv

VLLM_PRECOMPILED_WHEEL_LOCATION="https://github.com/vllm-project/vllm/releases/download/v0.22.0/vllm-0.22.0%2Bcu129-cp38-abi3-manylinux_2_28_x86_64.whl" \
VLLM_USE_PRECOMPILED=1 \
uv pip install --reinstall-package vllm --torch-backend=cu129 \
  "git+https://github.com/UpstageAI/[email protected]"

Start the server:

vllm serve upstage/Solar-Open2-250B \
  --served-model-name solar-open2-250b \
  --tensor-parallel-size 8 \
  --enable-expert-parallel \
  --moe-backend triton \
  --default-chat-template-kwargs '{"think_render_option":"preserved"}' \
  --reasoning-parser solar_open2 \
  --tool-call-parser solar_open2 \
  --enable-auto-tool-choice \
  --logits-processors vllm.v1.sample.logits_processor.solar_open2:SolarOpen2TemplateLogitsProcessor

Send a chat completion request:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "solar-open2-250b",
    "messages": [
      {"role": "user", "content": "What is Upstage?"}
    ],
    "max_tokens": 131584,
    "temperature": 1.0,
    "top_p": 1.0,
    "reasoning_effort": "high"
  }'

Quantized Versions

Official quantized models by NotaAI are available for deployment on smaller GPU configurations:

• Solar-Open2-250B-Nota-INT4-GlobalPruned
• Solar-Open2-250B-Nota-NVFP4
• Solar-Open2-250B-Nota-INT4

---

Capabilities

Reasoning

Use reasoning_effort="high" for reasoning and reasoning_effort="none" for a direct response. The recommended vLLM configuration limits a reasoning block to 131,072 tokens.

max_tokens limits the complete response, including reasoning and the final answer, so leave room beyond the reasoning cap.

from openai import OpenAI

client = OpenAI(api_key="EMPTY", base_url="http://localhost:8000/v1")

response = client.chat.completions.create(
    model="solar-open2-250b",
    messages=[
        {
            "role": "user",
            "content": "Prove that the square root of 2 is irrational.",
        },
    ],
    reasoning_effort="high",
    temperature=1.0,
    top_p=1.0,
    max_tokens=131584,
)

# The reasoning trace is returned separately from the final answer.
print(response.choices[0].message.reasoning)
print(response.choices[0].message.content)

Tool Calling

Tool calls follow the standard OpenAI function-calling interface. Start the server with --tool-call-parser solar_open2 and --enable-auto-tool-choice.

from openai import OpenAI

client = OpenAI(api_key="EMPTY", base_url="http://localhost:8000/v1")

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

response = client.chat.completions.create(
    model="solar-open2-250b",
    messages=[
        {
            "role": "user",
            "content": "What's the weather in Seoul?",
        },
    ],
    tools=tools,
)

print(response.choices[0].message.tool_calls)

Agentic Use

Both Anthropic's Claude Code and Nous Research's Hermes Agent can run on Solar Open 2 served locally with vLLM (see the vLLM deployment guide). A single vLLM server exposes both interfaces: Claude Code connects through the Anthropic-compatible /v1/messages endpoint and Hermes Agent through the OpenAI-compatible /v1 endpoint (model id solar-open2-250b), each needing only a few environment variables or one provider entry — no setup script required. Tools exposed over the Model Context Protocol (MCP) reach the model through the same tool-calling interface, and both agents support MCP natively.

Claude Code

vLLM exposes an Anthropic-compatible /v1/messages endpoint, so Claude Code connects directly — no proxy needed:

export ANTHROPIC_BASE_URL=http://localhost:8000
export ANTHROPIC_AUTH_TOKEN=dummy   # any non-empty value
export ANTHROPIC_MODEL=solar-open2-250b
export ANTHROPIC_SMALL_FAST_MODEL=solar-open2-250b
claude

The model name must match the server's --served-model-name (solar-open2-250b).

Prerequisites: the Claude Code CLI installed and a running vLLM server.

Hermes Agent

Register the local vLLM server as a custom OpenAI-compatible provider in ~/.hermes/config.yaml:

model:
  provider: custom
  default: solar-open2-250b
  base_url: http://localhost:8000/v1
  api_key: dummy

---

Best Practices

Recommended client-side generation settings (the values a client / API caller should send)

Solar Open 2 is a reasoning-capable model. Use reasoning_effort="high" for complex or agentic tasks. The recommended vLLM configuration preserves the reasoning trace.

Recommended settings by reasoning mode

• Set max_tokens high enough (up to 256K) — reasoning traces can be long and may otherwise truncate the answer.

• The reasoning trace is preserved by default (think_render_option=preserved).

• Multi-turn: Keep prior reasoning traces in the conversation history. The default think_render_option=preserved handles this automatically — do not strip reasoning from previous turns when constructing follow-up requests.

• Parsing: the OpenAI-compatible server returns reasoning in a separate message.reasoning field with local transformers, split the raw output on the reasoning markers yourself.

---

License

Solar Open 2 is distributed under the Upstage Solar License.

Key requirements for Derivative AI Models (create / train / fine-tune / distill / improve using Solar Open 2):

• Naming: prefix your model name with "Solar" (e.g., Solar-MyModel-v1).

• Attribution: prominently display "Built with Solar" in related public-facing materials.

• Notice: include a copy of the Upstage Solar License with your derivative model.

---

Citation

@misc{solar-open-2-2026,
    title={Solar Open 2 Technical Report},
    author={Upstage AI},
    year={2026},
    url={https://huggingface.co/upstage/Solar-Open2-250B}
}

---