gpt-oss-20b (free)
OpenAI · released Aug 5, 2025
gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for...
Specification
- Context window
- 131K
- Max output
- 33K
- Knowledge cutoff
- Jun 30, 2024
- Parameters
- 21.5B
- Licence
- apache-2.0
- Serving providers
- 1
- Moderated
- No
- Uptime
- 97.1%
Intelligence
15.2
49th percentile
Coding
20.7
Coding Index
Agentic
3.1
Agentic Index
Output speed
18 t/s
Median across providers
Latency
3.14s
Time to first token
Cost per task
Free
Estimated
Benchmarks
Where the score comes from
The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.
Evaluation scores
Percentage correct · higher is better
- AIME 202589.3%
- LiveCodeBench77.7%
- MMLU-Pro74.8%
- GPQA Diamond68.8%
- IFBench65.1%
- τ²-bench (Telecom)60.2%
- SciCode34.4%
- AA-LCR (long context)33.3%
- Humanity's Last Exam11.0%
- Terminal-Bench Hard10.6%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| AIME 2025 | 89.3% |
| LiveCodeBench | 77.7% |
| MMLU-Pro | 74.8% |
| GPQA Diamond | 68.8% |
| IFBench | 65.1% |
| τ²-bench (Telecom) | 60.2% |
| SciCode | 34.4% |
| AA-LCR (long context) | 33.3% |
| Humanity's Last Exam | 11.0% |
| Terminal-Bench Hard | 10.6% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- o3 Mini HighOpenAI15.7
- gpt-oss-20b (free)OpenAI15.2
- gpt-oss-20bOpenAI15.2
- 15.0
- GPT-4.1 MiniOpenAI14.8
- Mistral Medium 3.1Mistral AI14.7
- Solar Pro 3Upstage14.5
- Llama 4 MaverickMeta14.5
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| o3 Mini High | 15.7 |
| gpt-oss-20b (free) | 15.2 |
| gpt-oss-20b | 15.2 |
| Nemotron 3 Nano Omni (free) | 15.0 |
| GPT-4.1 Mini | 14.8 |
| Mistral Medium 3.1 | 14.7 |
| Solar Pro 3 | 14.5 |
| Llama 4 Maverick | 14.5 |
Percentile among all indexed models
Pricing
What it costs to run
List prices per million tokens, plus what one representative task works out to.
List price
- Input / 1M tokens
- Free
- Output / 1M tokens
- Free
- Cached input / 1M
- Not offered
- Blended 3:1
- Free
One task, estimated
Free
- Input tokens
- 50,000
- Output tokens
- 80,000
- Profile
- Reasoning
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).
Serving providers
Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.
Arena
Head-to-head generation quality
Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.
- Overall Elo
- 961
- Win rate
- 39.7%
- Strongest at
- Data visualisation
- Tournaments
- 78
Elo by category
Dot position on a 875–975 scale · Elo has no meaningful zero
- 961
- 875
Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.
View as table
| Category | Elo | Win rate |
|---|---|---|
| Data visualisation | 961 | 39.7% |
| Websites | 875 | 27.9% |