gpt-oss-120b
OpenAI · released Aug 5, 2025
gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases. It activates 5.1B parameters per forward pass and is optimized...
Specification
- Context window
- 131K
- Max output
- 66K
- Knowledge cutoff
- Jun 30, 2024
- Parameters
- 116.8B
- Licence
- apache-2.0
- Serving providers
- 21
- Moderated
- No
- Uptime
- 100.0%
Intelligence
11.6
48th percentile
Coding
30.4
Coding Index
Agentic
3.7
Agentic Index
Output speed
641 t/s
Median across providers
Latency
250ms
Time to first token
Cost per task
$0.06
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 202593.4%
- LiveCodeBench87.8%
- MMLU-Pro80.8%
- GPQA Diamond78.2%
- IFBench69.0%
- τ²-bench (Telecom)65.8%
- AA-LCR (long context)52.0%
- SciCode34.0%
- Terminal-Bench Hard23.5%
- Humanity's Last Exam19.6%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| AIME 2025 | 93.4% |
| LiveCodeBench | 87.8% |
| MMLU-Pro | 80.8% |
| GPQA Diamond | 78.2% |
| IFBench | 69.0% |
| τ²-bench (Telecom) | 65.8% |
| AA-LCR (long context) | 52.0% |
| SciCode | 34.0% |
| Terminal-Bench Hard | 23.5% |
| Humanity's Last Exam | 19.6% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- Mercury 2.5Inception12.3
- Sonar Reasoning ProPerplexity11.8
- gpt-oss-120bOpenAI11.6
- Qwen3.5-9BQwen11.2
- Granite 4.2 8BIBM Granite11.1
- GLM 4.5 AirZ.ai11.1
- 11.0
- o3 Mini HighOpenAI11.0
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| Mercury 2.5 | 12.3 |
| Sonar Reasoning Pro | 11.8 |
| gpt-oss-120b | 11.6 |
| Qwen3.5-9B | 11.2 |
| Granite 4.2 8B | 11.1 |
| GLM 4.5 Air | 11.1 |
| Qwen3 VL 32B Instruct | 11.0 |
| o3 Mini High | 11.0 |
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
- $0.15
- Output / 1M tokens
- $0.6
- Cached input / 1M
- $0.075
- Blended 3:1
- $0.262
One task, estimated
$0.06
- 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
- Venice
- CoreWeave
- DekaLLM
- AkashML
- DeepInfra
- Crusoe
- Mancer 2
- Novita
- DigitalOcean
- BaseTen
- Parasail
- SambaNova
- Amazon Bedrock
- Nebius
- SiliconFlow
- Phala
- Together
- Groq
- Mara
- Cerebras
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
- 995
- Win rate
- 40.5%
- Strongest at
- Game development
- Tournaments
- 445
Elo by category
Dot position on a 900–1,000 scale · Elo has no meaningful zero
- 995
- 988
- 970
- 928
- 902
Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.
View as table
| Category | Elo | Win rate |
|---|---|---|
| Game development | 995 | 40.5% |
| Data visualisation | 988 | 43.6% |
| Websites | 970 | 32.5% |
| UI components | 928 | 35.7% |
| 3D scenes | 902 | 29.4% |