OpenAIgpt-oss-20b
Best configuration ranks #355 of 610 on the BenchLeader Index at 47.1 ±4.6. Last measured 2 Sept 2026. Released 5 Aug 2025.
- Blended price
- $0.085/M
- $0.050 in · $0.190 out
- Output speed
- 192 tok/s
- First answer
- 11 s
- first token 0.43 s
- Context
- 131k
- Overall index47
- Reasoning49
- Coding55
- Agents & tools29
- Maths55
- Knowledge37
- Instruction following64
- Human preference48
- Long context43
- Composite40
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 | 44.6 | #411 | 249 tok/s | 8.87 s | $0.0001 | Agents & tools 38 · Composite 42 · Instruction following 58 · Knowledge 36 · Long context 41 · Maths 42 · Reasoning 40 |
| medium | – | – | 192 tok/s | 11 s | – | Coding 41 · Maths 53 · Reasoning 45 |
| high | – | – | 192 tok/s | 11 s | – | Coding 40 · Maths 46 · Reasoning 36 |
| defaultbest | 47.1 | #355 | 192 tok/s | 11 s | $0.0001 | Agents & tools 29 · Coding 55 · Composite 40 · Human preference 48 · Instruction following 64 · Knowledge 37 · Long context 43 · Maths 55 · Reasoning 49 |
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 | 53.2%#194 | 60.8%#177 | 46.0%#217 | – | Epoch AI Benchmarking Hub |
| LMArena Hard Prompts | – | – | – | 1322#219 | LMArena |
| GPQA Diamond (AA)not in index | 61.1%#350 | – | – | 68.8%#296 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 5.3%#361 | – | – | 11.0%#240 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | – | – | – | 68.9%#94 | Vals AI |
| Kagi LLM Benchmark | – | – | – | 53.2%#71 | Kagi LLM Benchmark |
Coding
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| SciCode | – | – | 34.4%#136 | – | SciCode |
| WeirdML | – | 36.8%#118 | 40.9%#99 | – | WeirdML |
| LMArena Coding | – | – | – | 1369#200 | LMArena |
| SciCode (AA)not in index | – | – | – | 38.9%#121 | Artificial Analysis |
| LiveCodeBench | – | – | – | 80.4%#67 | Vals AI |
Agents & tools
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| Terminal-Bench | – | – | – | 3.4%#66 | Terminal-Bench |
| Terminal-Bench Hard | 4.5%#270 | – | – | 10.6%#213 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 50.3%#190 | – | – | 60.2%#174 | Artificial Analysis |
Maths
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| OTIS Mock AIME | 40.3%#179 | 65.3%#138 | 50.8%#166 | – | Epoch AI Benchmarking Hub |
| AIME (Vals) | – | – | – | 86.0%#33 | Vals AI |
| MGSM | – | – | – | 89.0%#48 | Vals AI |
Knowledge
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| AA-Omniscience | -58.5#378 | – | – | -63.0#410 | Artificial Analysis |
| MMLU-Pro | – | – | – | 71.6%#111 | Vals AI |
| LegalBench | – | – | – | 70.8%#112 | Vals AI |
| CorpFin | – | – | – | 53.1%#95 | Vals AI |
| TaxEval | – | – | – | 63.7%#115 | Vals AI |
| MedQA | – | – | – | 82.9%#63 | Vals AI |
| MultiNRC | – | – | – | 10.4%#40 | Scale AI SEAL |
Instruction following
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| IFBench | 57.8%#133 | – | – | 65.1%#103 | Artificial Analysis |
Human preference
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| LMArena Text | – | – | – | 1317#218 | LMArena |
Long context
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| AA-LCR | 31.0%#322 | – | – | 34.7%#305 | Artificial Analysis |
Composite
| Benchmark | low | medium | high | default | Source |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | – | 137.8#107 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 9.9#313 | – | – | 9.0#338 | 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 |
|---|---|---|---|---|---|---|
| Groq | 384 tok/s | 0.54 s | $0.075 | $0.300 | 131k | – |
| Amazon Bedrock | 381 tok/s | 0.32 s | $0.070 | $0.150 | 131k | – |
| Google Vertex | 259 tok/s | 0.38 s | $0.070 | $0.250 | 131k | – |
| CoreWeave | 104 tok/s | 0.11 s | $0.030 | $0.130 | 131k | fp4 |
| DeepInfra | 85 tok/s | 0.31 s | $0.030 | $0.140 | 131k | bf16 |
| NovitaAI | 77 tok/s | 0.78 s | $0.040 | $0.150 | 131k | fp4 |
| Parasail | 66 tok/s | 0.48 s | $0.030 | $0.150 | 131k | fp4 |
| Phala | 64 tok/s | 0.28 s | $0.040 | $0.150 | 131k | – |
| SiliconFlow | 55 tok/s | 1.07 s | $0.040 | $0.180 | 131k | fp8 |
| Together | 48 tok/s | 0.30 s | $0.050 | $0.200 | 131k | – |
| Darkbloom | 28 tok/s | 3.28 s | $0.020 | $0.100 | 131k | fp8 |
| AkashML | 28 tok/s | 1.56 s | $0.020 | $0.100 | 131k | fp4 |
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 $0.037 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0001 | $0.0001 | 12.8 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0007 | $0.0006 | 14.3 s |
| Code edit | 6,000 / 1,500 | $0.0006 | $0.0005 | 19.0 s |
| Agentic coding session | 60,000 / 4,000 | $0.0038 | $0.0032 | 32.0 s |
| Structured extraction | 2,000 / 200 | $0.0001 | $0.0001 | 12.2 s |
See also
Data as of 9 Sept 2026. Compare these configurations.