OpenAIgpt-oss-120b
Best configuration ranks #283 of 610 on the BenchLeader Index at 50.5 ±7.7 (high reasoning effort). Last measured 5 Aug 2025stale: no new result in six months. Released 5 Aug 2025.
- Blended price
- –
- Output speed
- 202 tok/s
- First answer
- 11 s
- first token 0.49 s
- Context
- 128k
- Overall index51
- Reasoning56
- Coding46
- Maths51
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| low | 46.0 | #376 | 224 tok/s | 9.68 s | $0.0002 | Agents & tools 38 · Coding 39 · Composite 42 · Instruction following 58 · Knowledge 39 · Long context 49 |
| medium | – | – | 202 tok/s | 11 s | – | Coding 45 |
| highbest | 50.5 | #283 | 202 tok/s | 11 s | – | Coding 46 · Maths 51 · Reasoning 56 |
| default | 47.6 | #348 | 202 tok/s | 11 s | $0.0002 | Agents & tools 37 · Coding 43 · Composite 45 · Human preference 53 · Instruction following 51 · Knowledge 46 · Long context 52 · Maths 60 · Reasoning 41 |
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| GPQA Diamond | – | – | 75.8%#131 | – | Epoch AI Benchmarking Hub |
| SimpleBench | – | – | – | 22.1%#78 | SimpleBench |
| LMArena Hard Prompts | – | – | – | 1362#181 | LMArena |
| GPQA Diamond (AA)not in index | 67.2%#307 | – | – | 78.2%#207 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 5.9%#338 | – | – | 19.6%#162 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | – | – | – | 78.5%#70 | Vals AI |
| Kagi LLM Benchmark | – | – | – | 58.6%#53 | Kagi LLM Benchmark |
Coding
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| SciCode | 36.0%#127 | – | 38.9%#116 | – | SciCode |
| WeirdML | – | 41.9%#93 | 48.2%#70 | – | WeirdML |
| LMArena Coding | – | – | – | 1391#180 | LMArena |
| SciCode (AA)not in index | – | – | – | 34.0%#138 | Artificial Analysis |
| LiveCodeBench | – | – | – | 83.2%#49 | Vals AI |
| SWE-bench (Vals)not in index | – | – | – | 33.6%#83 | Vals AI |
| SWE-Bench Pro | – | – | – | 16.2%#19 | Scale AI SEAL |
| Aider Polyglot | – | – | – | 41.8%#28 | Aider polyglot leaderboard |
| SWE-bench Verified (bash only) | – | – | – | 26.0%#38 | SWE-bench |
| SWE-bench Verified (any scaffold)not in index | – | – | – | 26.0%#57 | SWE-bench |
Agents & tools
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| Terminal-Bench | – | – | – | 18.7%#57 | Terminal-Bench |
| APEX-Agents | – | – | – | 4.7%#57 | Mercor |
| Terminal-Bench Hard | 5.3%#263 | – | – | 23.5%#143 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 45.0%#206 | – | – | 65.8%#161 | Artificial Analysis |
Maths
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| OTIS Mock AIME | – | – | 88.9%#61 | – | Epoch AI Benchmarking Hub |
| AIME (Vals) | – | – | – | 92.6%#18 | Vals AI |
| MGSM | – | – | – | 92.0%#24 | Vals AI |
| MathArena Apex | – | – | 1.0%#35 | – | MathArena |
Knowledge
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| AA-Omniscience | -53.5#347 | – | – | -49.3#311 | Artificial Analysis |
| MMLU-Pro | – | – | – | 79.2%#94 | Vals AI |
| LegalBench | – | – | – | 75.9%#103 | Vals AI |
| CorpFin | – | – | – | 58.2%#82 | Vals AI |
| TaxEval | – | – | – | 71.6%#74 | Vals AI |
| MedQA | – | – | – | 91.4%#38 | Vals AI |
| PRBench Finance | – | – | – | 43.8%#18 | Scale AI SEAL |
| PRBench Legal | – | – | – | 40.2%#24 | Scale AI SEAL |
| MultiNRC | – | – | – | 15.2%#38 | Scale AI SEAL |
Instruction following
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| IFBench | 58.3%#129 | – | – | 69.0%#75 | Artificial Analysis |
| MultiChallenge | – | – | – | 45.3%#24 | Scale AI SEAL |
Human preference
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| LMArena Text | – | – | – | 1352#178 | LMArena |
Long context
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| AA-LCR | 46.0%#264 | – | – | 52.0%#250 | Artificial Analysis |
Composite
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | – | 140.1#99 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 10.2#309 | – | – | 12.3#268 | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Cerebras | 696 tok/s | 0.19 s | $0.350 | $0.750 | 131k | fp16 |
| Amazon Bedrock | 293 tok/s | 0.41 s | $0.150 | $0.600 | 131k | – |
| SambaNova | 256 tok/s | 0.68 s | $0.140 | $0.950 | 131k | – |
| Groq | 250 tok/s | 0.27 s | $0.150 | $0.600 | 131k | – |
| DeepInfra (fp8) | 166 tok/s | 1.94 s | $0.200 | $0.950 | 131k | fp8 |
| Baseten (US) | 157 tok/s | 0.30 s | $0.100 | $0.500 | 128k | fp4 |
| Baseten | 155 tok/s | 0.31 s | $0.100 | $0.500 | 128k | fp4 |
| Google Vertex | 140 tok/s | 0.34 s | $0.090 | $0.360 | 131k | – |
| Nebius Token Factory | 140 tok/s | 0.54 s | $0.150 | $0.600 | 131k | fp4 |
| Amazon Bedrock (EU) | 132 tok/s | 0.43 s | $0.150 | $0.600 | 131k | – |
| DeepInfra (Turbo) | 121 tok/s | 0.42 s | $0.150 | $0.600 | 131k | bf16 |
| MARA | 94 tok/s | 3.02 s | $0.150 | $0.750 | 131k | – |
| Parasail | 91 tok/s | 0.44 s | $0.100 | $0.750 | 131k | fp4 |
| Phala | 71 tok/s | 1.19 s | $0.150 | $0.600 | 131k | – |
| Together | 64 tok/s | 0.30 s | $0.150 | $0.600 | 131k | – |
| NovitaAI | 61 tok/s | 1.17 s | $0.050 | $0.250 | 131k | fp4 |
| AkashML | 47 tok/s | 1.32 s | $0.030 | $0.170 | 131k | bf16 |
| Mancer | 36 tok/s | 0.61 s | $0.055 | $0.500 | 131k | fp8 |
| CoreWeave | 33 tok/s | 0.45 s | $0.030 | $0.170 | 131k | fp4 |
| DigitalOcean | 23 tok/s | 0.75 s | $0.055 | $0.385 | 128k | – |
| DeepInfra (bf16) | 22 tok/s | 2.67 s | $0.037 | $0.170 | 131k | bf16 |
| SiliconFlow | 10 tok/s | 5.42 s | $0.050 | $0.450 | 131k | fp8 |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.075 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | – | – | 12.2 s |
| Summarise a 30-page report | 12,000 / 600 | – | – | 13.7 s |
| Code edit | 6,000 / 1,500 | – | – | 18.1 s |
| Agentic coding session | 60,000 / 4,000 | – | – | 30.5 s |
| Structured extraction | 2,000 / 200 | – | – | 11.7 s |
See also
Data as of 9 Sept 2026. Compare these configurations.