# GPT-OSS-120b

> OpenAI gpt-oss-120b open-weight MoE via SiliconFlow free tier — reasoning and tool-use support.

**ID**: `openai/gpt-oss-120b`  
**Creator**: OpenAI  
**Category**: text  
**Context**: 65K tokens  
**Pricing**: $0.04 / $0.17 per 1M · billed by your provider (BYOK)  
**Released**: 2026-07-08  
**Web page**: https://anyrouter.dev/model/openai/gpt-oss-120b

**Input modalities**: text  
**Output modalities**: text  

**Capabilities**: chat, reasoning, streaming, function-calling

## Usage

**Endpoint**: `POST https://anyrouter.dev/api/v1/chat/completions`

```bash
curl https://anyrouter.dev/api/v1/chat/completions \
  -H "Authorization: Bearer $ANYROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "openai/gpt-oss-120b",
  "messages": [
    {
      "role": "user",
      "content": "Say hi in 3 words."
    }
  ]
}'
```

## Providers

| Provider | Input | Output |
| --- | --- | --- |
| fpt-byok | $0.14/M (list) | $0.61/M (list) |
| siliconflow-byok | $0.14/M (list) | $0.60/M (list) |
| cline-byok | $0.14/M (list) | $0.60/M (list) |
| cerebras-byok | $0.35/M (list) | $0.75/M (list) |
| ovhcloud-byok | $0.09/M (list) | $0.47/M (list) |
| sambanova-byok | $0.14/M (list) | $0.60/M (list) |
| router-byok | $0.15/M (list) | $0.60/M (list) |
| vercel-byok | $0.10/M (list) | $0.50/M (list) |
| openrouter-byok | $0.04/M (list) | $0.17/M (list) |
| deepinfra-byok | $0.04/M (list) | $0.17/M (list) |
| novita-byok | $0.05/M (list) | $0.25/M (list) |
| gmicloud-byok | $0.05/M (list) | $0.25/M (list) |
| huggingface-byok | $0.14/M (list) | $0.60/M (list) |

BYOK routes are billed by your own provider key — AnyRouter charges $0. Rates marked (list) are the provider's published list price.

## Benchmarks

- **GPQA Diamond**: 0.7
- **HLE**: 0.1
- **τ²-Bench Telecom**: 0.5
- **Aider Polyglot**: 0.4
- **SciCode**: 0.4

Benchmark data from [Aider](https://aider.chat/docs/leaderboards/) (Apache-2.0).

## Model card

- Architecture: mixture of experts [3]
- Total parameters: 116.8B [3,1]
- Active parameters: 5.1B [3,1]
- Native context: 131072 tokens [3,2]
- Training cutoff: 2024-06 [3]
- License: Apache-2.0, open weights (https://huggingface.co/openai/gpt-oss-120b) [1]

### Strengths [1]

- Runs on one 80 GB GPU with MXFP4 weights
- Reasoning effort set to low, medium or high
- Native function calling, web browsing and Python execution

### Known limits [3,1]

- Hallucinates more than o4-mini on SimpleQA and PersonQA
- Raw chain of thought can hallucinate; filter it before showing users
- Works only with the harmony response format

### Use cases [1]

- General-purpose, high-reasoning production workloads
- Agentic tasks with browsing and function calls
- Fine-tuning on a single H100 node

### API notes [1]

- Reasoning level set in the system prompt, e.g. "Reasoning: high"
- Supports function calling and structured outputs

### Links

- announcement: https://openai.com/index/introducing-gpt-oss/
- model card: https://huggingface.co/openai/gpt-oss-120b
- paper: https://arxiv.org/abs/2508.10925

### Sources

1. [openai/gpt-oss-120b model card](https://huggingface.co/openai/gpt-oss-120b/raw/main/README.md) (accessed 2026-10-14)
2. [gpt-oss-120b config.json](https://huggingface.co/openai/gpt-oss-120b/raw/main/config.json) (accessed 2026-10-14)
3. [gpt-oss-120b & gpt-oss-20b Model Card](https://arxiv.org/abs/2508.10925) (accessed 2026-10-14)

## Source

- Upstream docs: https://docs.siliconflow.com