This model is 15 months old. A newer version of it has since shipped: GPT-6 Cyber (2026-11-05). The details below are still accurate for this model. It just is not what we would recommend today.
gpt-oss-20b
OpenAIπΊπΈ United States
The smaller of OpenAI's two gpt-oss open-weights mixture-of-experts models, Apache 2.0, sized for local and edge use.
Context window131K tokens
Input / 1M tokens$0.03
Output / 1M tokens$0.13
Pricing and context window via OpenRouter, 2026-11-08.
Version History
launchmajor
gpt-oss-20b released. The smaller of OpenAI's two gpt-oss open-weights mixture-of-experts models, Apache 2.0, sized for local and edge use.
Benchmark Scores
Full leaderboard β| Benchmark | Score | Rank | Source |
|---|---|---|---|
| Arena Elo | 1317 Elo | 136 of 151 | LMArena, Nov 2026 β |
| GPQA | 68.8% | 137 of 159 | Artificial Analysis, Jul 2026 β |
| Speed (tok/s) | 192 tok/s | 22 of 127 | Artificial Analysis, Aug 2026 β |
| Hallucination Rate (lower is better) | 86.7% | 87 of 109 | Artificial Analysis, Aug 2026 β |
| MMLU-Pro | 74.8% | 58 of 73 | Artificial Analysis, Jul 2026 β |
| LiveCodeBench | 77.7% | 24 of 48 | Artificial Analysis, Jul 2026 β |
Rank is against every model in our database with a sourced score on that benchmark, each counted once with its best score.