Models · OpenAI · Out sinceReleased 5 Aug 2025
gpt-oss-20b
gpt-oss-20b is made by OpenAI. We don't have enough test results yet to rank it. It's cheap to use. Its makers have published it, so you can run it on your own servers.
21B total / 3.6B active MoE, native MXFP4 MoE weights, runs within 16GB memory (model card). Apache 2.0 + gpt-oss usage policy. Still OpenAI's newest general open-weight LLM as of 2026-10-01 (only gpt-oss-safeguard and privacy-filter released since). No OpenAI first-party API price (OpenRouter ~$0.018/$0.09).
- —
- —
- —
- Cheap$0.02 / $0.09
- $0.04
- 131K
- —
- 9 (9 independent9 indep.)
- 5 Aug 2025
- OpenAI
- text
Scores are out of 100 for each area, compared with every model we track; “#” is its rank. Context is how much text it can read at once (1M is roughly 700,000 words). Blended price mixes the cost of what you send and what it writes back.
OpenRouter is a service that gives access to many AI models in one place. This model is listed there as openai/gpt-oss-20b.
Route it as openai/gpt-oss-20b at $0.02 in / $0.09 out per 1M tokens, 131K context. Listed since 5 Aug 2025.
frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokensmin_ppresence_penaltyreasoningreasoning_effortrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p
Run it yourself
Self-hosting
On your own servers,Licence, sizeunder your own control.and hardware.
- Yes
- 21B / 3.6B
- No
Runs on a powerful laptop or workstation.
≤ 24 GB at 4-bit (one consumer GPU / 32 GB Mac). Weights ≈ 22 GB at 8-bit, 12 GB at 4-bit (+10–30% for KV cache). MoE: 3.6B active per token.
Licence conditions: Usage policy: comply with applicable law.
Restrictions: Usage policy: comply with applicable law.
Languages: Languages: Not stated.
Training it further: Fine-tuning: Card: can be fine-tuned on consumer hardware.
Quantisations: mxfp4 (native), gguf (ggml-org, unsloth). Engines: Transformers, vLLM, Ollama, LM Studio.
Download (Hugging Face)Weights on Hugging Face ↗huggingface.co ↗
Every test result
Every result
All the numbers,with where they came from.
Every result we've found for this model, with who measured it and a link to where we read it.
All raw rows (vendor and independent kept separate), with setting label, source, date, harness and notes.
| NotesNotes | ||||||
|---|---|---|---|---|---|---|
| AA-LCR | 34.7% | Highgpt-oss-20b (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug gpt-oss-20b |
| AA-Omniscience Index | -63.1 | Highgpt-oss-20b (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug gpt-oss-20b; AA-Omniscience Index (-100..100) |
| Artificial Analysis Intelligence Index | 9 | Highgpt-oss-20b (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug gpt-oss-20b |
| Artificial Analysis output speed | 174 tok/s | Highgpt-oss-20b (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug gpt-oss-20b; median output tokens/sec across AA-tracked API providers (not a self-hosted measurement) |
| EuroEval Swedish (generative) | 2 | Mediumopenai/gpt-oss-20b#medium | Independent testIndependentEuroEval (Alexandra Institute) ↗ | 29 Sep 2026 | — | rank tier 9; ±0.18; lower is better; #low 2.12, #high 2.48; SweDN 34.52, Skolprov 74.89, Swedish facts 23.93, ScaLA-sv 52.42 |
| GPQA Diamond | 68.8% | Highgpt-oss-20b (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug gpt-oss-20b |
| Humanity's Last Exam | 11.0% | Highgpt-oss-20b (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug gpt-oss-20b |
| SciCode | 38.9% | Highgpt-oss-20b (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug gpt-oss-20b |
| Terminal-Bench 2.1 | 13.9% | Highgpt-oss-20b (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug gpt-oss-20b |