gpt-oss

weights

open(gpt-oss-120b/20b on HF, ungated safetensors)

data

closed

code

partial(inference reference implementations (PyTorch/Triton) in openai/gpt-oss

license

Apache-2.0(OSI)

gpt-oss ships as Apache-2.0 weights, downloadable without a gate in both the 120B and 20B sizes. The training data is not released, and the repository holds only inference reference code rather than a training pipeline, which keeps it at open weights rather than a more open score.

• https://api.github.com/repos/openai/gpt-oss recorded 2026-09-01

  `license: Apache-2.0`, 20,362 stars, description "gpt-oss-120b and gpt-oss-20b are two open-weight language models by OpenAI"; the repo is reference implementations and tooling, not a training pipeline, which holds `code` at partial.

• https://arxiv.org/abs/2508.10925 recorded 2026-10-07

  The gpt-oss-120b & gpt-oss-20b Model Card paper (OpenAI, 2025). Describes training at a high level; no corpus or pipeline release accompanies it. Corroborates data:closed.

• https://huggingface.co/api/models/openai/gpt-oss-120b recorded 2026-10-07

  Model API for openai/gpt-oss-120b: `gated: false`, `private: false`, cardData.license `apache-2.0`, safetensors weights listed.

• https://huggingface.co/openai/gpt-oss-120b/raw/main/README.md recorded 2026-10-07

  Raw card front matter records only `license: apache-2.0` with no datasets key; the body markets "Permissive Apache 2.0 license", configurable reasoning effort and fine-tunability, and carries no training-data section — the recipe is not released.

• https://raw.githubusercontent.com/openai/gpt-oss/main/README.md recorded 2026-10-07

  README: "This repository provides a collection of reference implementations" in PyTorch, Triton and Metal, "largely reference implementations for educational purposes and are not expected to be run in production", with inference examples and no training pipeline.

Adoption is measured as Hugging Face downloads summed across the two declared gpt-oss checkpoints, with the smaller 20B model drawing slightly more traffic than the 120B. Qwen and Gemma are the only other open-weight lines with comparable Hugging Face download volume.

• https://huggingface.co/api/models/openai/gpt-oss-120b recorded 2026-09-01

  5,376,994 downloads in the trailing 30 days for openai/gpt-oss-120b

• https://huggingface.co/api/models/openai/gpt-oss-20b recorded 2026-09-01

  6,519,052 downloads in the trailing 30 days for openai/gpt-oss-20b

gpt-oss-120b scores 24.1 on the Artificial Analysis Intelligence Index, clearly ahead of non-reasoning open models such as Llama 4 Maverick at 14.5, but well short of closed frontier models scoring near 60. That places it level with DeepSeek-R1, another open-weight reasoning line competitive on math and reasoning benchmarks without leading the category.

• https://artificialanalysis.ai/models/gpt-oss-120b recorded 2026-09-01

  AA model page for gpt-oss-120b (high): embedded index data reads artificialAnalysisIntelligenceIndex 24.1 against a leaderboard whose frontier entries read ~60 (Claude Opus 5 max 63.1, GPT-5.6 Sol 60.9, Kimi K3 59.7) and whose Llama 4 Maverick entry reads 14.5; page summary: "amongst the leading models in intelligence, but somewhat expensive when comparing to other open weight models of similar size".