GPT-OSS

gpt-oss is a pair of open-weight language models from OpenAI: gpt-oss-120b (117B total parameters, 5.1B active per token) and gpt-oss-20b (21B total, 3.6B active). Both are mixture-of-experts Transformers with a native context length of up to 128k tokens, trained on a mostly English, text-only dataset focused on STEM, coding and general knowledge. They are OpenAI's first open-weight language models since GPT-2.

The models support configurable reasoning effort (low, medium, high), full chain-of-thought output, function calling, web browsing and Python tool use, and Structured Outputs. OpenAI reports that gpt-oss-120b reaches near-parity with o4-mini on core reasoning benchmarks, and gpt-oss-20b is comparable to o3-mini. MoE weights were post-trained with MXFP4 quantization, so the 120b model runs on a single 80 GB GPU and the 20b model within 16 GB of memory.

Weights are downloadable from Hugging Face and can be run with Transformers, vLLM and other runtimes, or fine-tuned and hosted on-premises. The models must be used with OpenAI's harmony response format to work correctly. They are text-only (no image input), and the chain-of-thought is not intended to be shown to end users.