gpt-oss-20b

Facts

Released

2025-08-04 the day the repository was first published on Hugging Face Hugging Face, read 2026-09-24

Licence

apache-2.0 the licence the model card declares Hugging Face, read 2026-09-24

Open weights

Yes the weights are published in this Hugging Face repository Hugging Face, read 2026-09-24

Parameters

20.9 billion counted from the safetensors weight files Hugging Face, read 2026-09-24

Active parameters

4.2 billion estimate estimated from the expert counts and sizes in the configuration model configuration, read 2026-09-24

Knowledge cutoff

2024-06-30 the knowledge cutoff OpenRouter lists OpenRouter, read 2026-09-24

Context window

• 131,072 tokens the maximum position embeddings in the model configuration model configuration, read 2026-09-24
• 131,072 tokens the context length OpenRouter lists OpenRouter, read 2026-09-24

Longest output

• 32,768 tokens the largest output OpenRouter's first provider allows OpenRouter, read 2026-09-24

Tool calling

• Yes OpenRouter lists tools among the supported parameters OpenRouter, read 2026-09-24

Structured output

• Yes OpenRouter lists structured outputs among the supported parameters OpenRouter, read 2026-09-24

Reasoning controls

• Yes OpenRouter lists reasoning controls, with the efforts high, medium, low OpenRouter, read 2026-09-24

Inputs and outputs

text in, text out Hugging Face, read 2026-09-24

Good for

• Reasoning: OpenRouter lists reasoning controls for it OpenRouter, read 2026-09-24

Prices

Run it on your own hardware

Weights are the sizes of the files a source lists, or an estimate from the parameter count where none does. Memory is an estimate: weights plus KV cache plus 512 MiB and 5 percent of the weights for runtime buffers. Check it against your hardware.

Start it with a local runtime

• Ollama

  ollama run hf.co/unsloth/gpt-oss-20b-GGUF:Q4_0

• LM Studio

  lms get unsloth/gpt-oss-20b-GGUF
  lms server start

• llama.cpp server

  llama-server -hf unsloth/gpt-oss-20b-GGUF:Q4_0 --jinja

• vLLM

  vllm serve openai/gpt-oss-20b

• SGLang

  sglang serve --model-path openai/gpt-oss-20b --port 30000

Use it from your harness

Set up for DeepInfra with the model openai/gpt-oss-20b. Each endpoint page has the same setup for its own address.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "deepinfra": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "DeepInfra",
      "options": {
        "baseURL": "https://api.deepinfra.com/v1/openai",
        "apiKey": "{env:DEEPINFRA_TOKEN}"
      },
      "models": {
        "openai/gpt-oss-20b": {
          "name": "gpt-oss-20b",
          "limit": {
            "context": 131072,
            "output": 32768
          }
        }
      }
    }
  }
}

{
  "providers": {
    "deepinfra": {
      "baseUrl": "https://api.deepinfra.com/v1/openai",
      "api": "openai-completions",
      "apiKey": "$DEEPINFRA_TOKEN",
      "models": [
        {
          "id": "openai/gpt-oss-20b"
        }
      ]
    }
  }
}

export ANTHROPIC_BASE_URL="https://api.deepinfra.com/anthropic"
export ANTHROPIC_AUTH_TOKEN="$DEEPINFRA_TOKEN"
export ANTHROPIC_MODEL="openai/gpt-oss-20b"
claude

Other ways to reach it

• FastRouter, model openai/gpt-oss-20b
• Groq, model openai/gpt-oss-20b
• Hugging Face, model openai/gpt-oss-20b
• IO.NET, model openai/gpt-oss-20b
• Kilo Gateway, model openai/gpt-oss-20b
• NanoGPT, model openai/gpt-oss-20b
• NovitaAI, model openai/gpt-oss-20b
• Nvidia, model openai/gpt-oss-20b
• OpenRouter, model openai/gpt-oss-20b
• Pioneer, model openai/gpt-oss-20b
• SiliconFlow, model openai/gpt-oss-20b

Published results

These are results other people published. Baltor did not run them and does not rank models by them.

• Results the maker published on the model card published by openai Hugging Face, read 2026-09-24
• Artificial Analysis Intelligence Index: 9 published by Artificial Analysis OpenRouter, read 2026-09-24
• Artificial Analysis Coding Index: 20.7 published by Artificial Analysis OpenRouter, read 2026-09-24
• Artificial Analysis Agentic Index: 1.2 published by Artificial Analysis OpenRouter, read 2026-09-24