gpt-oss-120b
OpenAI · LLM model
gpt-oss-120b is OpenAI's largest open-weight language model, released in August 2025 under Apache 2.0. As of October 2026 it is still OpenAI's main open-weight LLM; it can be downloaded and run on a single high-end GPU.
Current Open weights
ReleasedAug 5, 2025
Context window131,072 tokens
Input → outputtext → text
API price—
Consumer accessFree download of weights (Apache 2.0) on Hugging Face; runs locally or via third-party hosts
Best for
Teams that need a capable reasoning model they can self-host, fine-tune and run privately on a single 80GB GPU.
Overview
gpt-oss-120b is OpenAI's largest open-weight language model, released in August 2025 under Apache 2.0. As of October 2026 it is still OpenAI's main open-weight LLM; it can be downloaded and run on a single high-end GPU.
Strengths and limits
- Apache 2.0 license allows commercial use and customization
- Mixture-of-experts: 117B total / 5.1B active parameters, fits on one 80GB GPU (H100 / MI300X)
- Adjustable reasoning level (low / medium / high)
- Text only, knowledge cutoff June 2024, well behind OpenAI's current hosted models
- Requires OpenAI's harmony response format to work correctly
Published benchmark scores
MMLU-Pro80.8
Source: huggingface.co
Scores are as published by the source; settings (reasoning effort, tools, attempts) differ between labs, so compare with care. How to read benchmarks →