gpt-oss-20b
OpenAI's 21B open-weight reasoning model under Apache 2.0, near o3-mini class on a single GPU.
About
Released August 5, 2025, gpt-oss-20b is the smaller of OpenAI's first open-weight models since GPT-2: a 21B-parameter mixture-of-experts (24 layers, 32 experts, 3.6B active per token) with adjustable reasoning effort, tool use, and instruction following. OpenAI reports it matching o3-mini on core reasoning benchmarks, including 71.5% on GPQA Diamond without tools, while fitting in 16GB of memory for local or edge deployment.
The weights are Apache 2.0 with a complementary acceptable-use policy, and the model ships in MXFP4 with broad ecosystem support (Ollama, llama.cpp, vLLM). It is text-only with chain-of-thought reasoning, so it will not handle images or audio, and as with any open reasoning model, downstream deployers own the safety tuning.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- GPQA Diamond (no tools): 71.5 (OpenAI-reported)
- Near-parity with o3-mini on core reasoning benchmarks (OpenAI-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: OpenAI's first open-weight release since GPT-2; millions of HF downloads within weeks; Apache 2.0 with complementary use policy; OpenAI-reported benchmarks near o3-mini; independent community testing broadly confirms competence. License: Apache 2.0 plus OpenAI usage policies.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · OpenAI's first open-weight release since GPT-2; millions of HF downloads within weeks
- · Apache 2.0 with complementary use policy
- · OpenAI-reported benchmarks near o3-mini; independent community testing broadly confirms competence