OpenAIgpt-oss-20b
gpt-oss-20b is an OpenAI open-weights reasoning model, released 5 Aug 2025. Its best configuration (high reasoning effort) ranks #540 of 785 on the BenchLeader Index at 44.3 ±4.8, in the lower half. The ± is the point: 312 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (65) and lowest in knowledge (33). At $0.093 per million tokens blended it is among the cheapest fifth of ranked models. Output speed of 159 tokens per second puts it in the fastest quarter, with a first answer in 13.4 s. It has been measured at 4 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 5 Aug 2025.stale: no new result in six months
- Blended price
- $0.093/M
- $0.050 in · $0.220 out
- Output speed
- 159 tok/s
- measured by Artificial Analysis
- First answer
- 13 s
- first token 0.29 s
- Context
- 131k
- Full answer
- 17 s
- median, reasoning included
- Cost per run
- $0.011
- one full Intelligence Index run
- Released
- 5 Aug 2025
- Overall index44
- Reasoning39
- Coding41
- Agents & tools37
- Maths46
- Knowledge33
- Instruction following65
- Long context42
- 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. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| low | 44.0 | #555 | 183 tok/s | 12 s | $0.0001 | Agents & tools 38 · Composite 41 · Instruction following 58 · Knowledge 35 · Long context 40 · Maths 41 · Reasoning 41 |
| medium | – | – | 115 tok/s | 0.29 s | $0.0001 | Coding 41 · Maths 52 · Reasoning 45 |
| highbest | 44.3 | #540 | 159 tok/s | 13 s | $0.0001 | Agents & tools 37 · Coding 41 · Composite 40 · Instruction following 65 · Knowledge 33 · Long context 42 · Maths 46 · Reasoning 39 |
| not stated | 44.1 | #547 | 115 tok/s | 0.29 s | $0.0001 | Agents & tools 20 · Coding 55 · Human preference 47 · Instruction following 36 · Knowledge 38 · Maths 57 · Reasoning 48 |
How its index has moved
9 Sept 2026 to 21 Oct 2026- high
- not stated
- low
The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. 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 | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| GPQA Diamond | 53.2%#209 | 60.8%#191 | 46.0%#234 | – | Epoch AI Benchmarking Hub | |
| Arena Hard Promptseffort not stated | – | – | – | 1323#240 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 61.1%#369 | – | 68.8%#309 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 5.3%#438 | – | 11.0%#304 | – | Artificial Analysis | |
| GPQA Diamond (Vals)not in indexeffort not stated | – | – | – | 68.9%#93 | Vals AI | |
| Kagi LLM Benchmarkeffort not stated | – | – | – | 53.2%#71 | Kagi LLM Benchmark | |
| GPQA Diamond (HELM)not in indexeffort not stated | – | – | – | 59.4%#28 | HELM Capabilities | |
| CritPt | 0.0%#315 | – | 1.4%#211 | – | Artificial Analysis | |
| LMCAnot in index | – | – | 14.5%#173 | – | Epoch AI Benchmarking Hub | |
| DTBenchnot in index | – | – | 68.0%#162 | – | Epoch AI Benchmarking Hub |
Coding
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| SciCode | – | 38.9%#171 | – | SciCode | |
| WeirdML | 36.8%#121 | 40.9%#101 | – | WeirdML | |
| Arena Codingeffort not stated | – | – | 1368#226 | Arena (LMArena) | |
| SciCode (AA)not in index | – | 38.9%#183 | – | Artificial Analysis | |
| LiveCodeBencheffort not stated | – | – | 80.4%#67 | Vals AI | |
| ALE-Benchnot in indexeffort not stated | – | – | 566.0#101 | Epoch AI Benchmarking Hub | |
| Terminal-Bench 4.0 (AA)not in index | – | 0.0%#152 | – | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | – | 13.9%#184 | – | Artificial Analysis |
Agents & tools
| Benchmark | low | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| Terminal-Bencheffort not stated | – | – | – | 3.4%#84 | Terminal-Bench | |
| Terminal-Bench Hard | 4.5%#284 | – | 10.6%#221 | – | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 50.3%#196 | – | 60.2%#179 | – | Artificial Analysis | |
| GDPval-AA v2.1not in index | – | – | 0.0%#124 | – | Artificial Analysis | |
| τ³-Banking (AA) | – | – | 7.0%#163 | – | Artificial Analysis | |
| APEX-Agents (AA)not in index | – | – | 0.7%#33 | – | Artificial Analysis | |
| AutomationBenchnot in index | – | – | 0.2%#219 | – | Artificial Analysis | |
| GDP.pdf (AA)not in index | – | – | 2.0%#190 | – | Artificial Analysis | |
| BrowseComp-Plusnot in index | 4.0%#20 | 16.9%#10 | 21.4%#8 | – | BrowseComp-Plus |
Maths
| Benchmark | low | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| OTIS Mock AIME | 40.3%#197 | 65.3%#154 | 50.8%#184 | – | Epoch AI Benchmarking Hub | |
| AIME (Vals)effort not stated | – | – | – | 86.0%#33 | Vals AI | |
| MGSMeffort not stated | – | – | – | 89.0%#47 | Vals AI | |
| Omni-MATH (HELM)effort not stated | – | – | – | 56.5%#11 | HELM Capabilities | |
| MATH 500not in indexeffort not stated | – | – | – | 94.2%#10 | Vals AI | |
| Arena Mathseffort not stated | – | – | – | 1335#208 | Arena (LMArena) |
Knowledge
| Benchmark | low | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| AA-Omniscience | -58.5#454 | -63.0#486 | – | Artificial Analysis | |
| MMLU-Proeffort not stated | – | – | 71.6%#110 | Vals AI | |
| LegalBencheffort not stated | – | – | 70.8%#117 | Vals AI | |
| CorpFineffort not stated | – | – | 53.1%#95 | Vals AI | |
| TaxEvaleffort not stated | – | – | 63.7%#115 | Vals AI | |
| MedQAeffort not stated | – | – | 82.9%#62 | Vals AI | |
| MultiNRCeffort not statedpossible contamination | – | – | 10.4%#41 | Scale AI SEAL | |
| MMLU-Pro (HELM)not in indexeffort not stated | – | – | 74.0%#31 | HELM Capabilities | |
| CaseLaw v2not in indexeffort not stated | – | – | 43.8%#52 | Vals AI | |
| AA-Omniscience: accuracynot in index | 15.1%#423 | 16.0%#401 | – | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 13.3%#363 | 5.9%#513 | – | Artificial Analysis |
Instruction following
| Benchmark | low | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| IFBench | 57.8%#136 | 65.1%#104 | – | Artificial Analysis | |
| IFEval (HELM)effort not stated | – | – | 73.2%#60 | HELM Capabilities | |
| WildBench (HELM)effort not stated | – | – | 73.7%#55 | HELM Capabilities | |
| Arena Instruction Followingnot in indexeffort not stated | – | – | 1281#257 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1317#240 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1240#273 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1289#251 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1303#254 | Arena (LMArena) |
Long context
| Benchmark | low | high | Source | Trend |
|---|---|---|---|---|
| AA-LCR | 31.0%#394 | 34.7%#377 | Artificial Analysis | |
| MLCRnot in index | – | 1.1%#165 | Artificial Analysis |
Composite
| Benchmark | low | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in indexeffort not stated | – | – | 137.7#123 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 9.9#384 | 9.0#417 | – | Artificial Analysis | |
| HELM Capabilities meannot in indexeffort not stated | – | – | 67.4%#28 | HELM Capabilities |
Safety & honesty
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| FORTRESSnot in index | 17.6%#44 | Scale AI SEAL | |
| MASKnot in indexpossible contamination | 86.5%#14 | Scale AI SEAL |
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 |
|---|---|---|---|---|---|---|
| Amazon Bedrock | 447 tok/s | 0.40 s | $0.070 | $0.150 | 131k | – |
| Groq | 425 tok/s | 0.61 s | $0.075 | $0.300 | 131k | – |
| CoreWeave | 117 tok/s | 0.33 s | $0.030 | $0.130 | 131k | fp4 |
| Parasail | 105 tok/s | 0.55 s | $0.030 | $0.150 | 131k | fp4 |
| AkashML | 93 tok/s | 0.52 s | $0.020 | $0.100 | 131k | fp4 |
| DeepInfra | 76 tok/s | 0.22 s | $0.030 | $0.140 | 131k | bf16 |
| SiliconFlow | 53 tok/s | 1.05 s | $0.040 | $0.180 | 131k | fp8 |
| DekaLLM | 52 tok/s | 0.51 s | $0.029 | $0.140 | 131k | bf16 |
| Darkbloom | 26 tok/s | 2.87 s | $0.018 | $0.090 | 131k | fp8 |
Price history
The price above, per 1M tokens, every day it changed.
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 | 15.3 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0007 | $0.0006 | 17.2 s |
| Code edit | 6,000 / 1,500 | $0.0006 | $0.0006 | 22.9 s |
| Agentic coding session | 60,000 / 4,000 | $0.0039 | $0.0033 | 38.6 s |
| Structured extraction | 2,000 / 200 | $0.0001 | $0.0001 | 14.7 s |
More from OpenAI
See also
Questions people ask
- What are gpt-oss-20b's benchmark scores?
- As of 24 Oct 2026, gpt-oss-20b (high effort) scores 44.3 ±4.8 on the BenchLeader Index, strongest in instruction following (65) and weakest in knowledge (33). That rests on 11 independent benchmark results, and every score on this page links to its publisher.
- How does gpt-oss-20b rank?
- gpt-oss-20b (high effort) is #540 of 785 ranked configurations on 24 Oct 2026. 312 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does gpt-oss-20b cost per 1M tokens?
- As of 24 Oct 2026, gpt-oss-20b costs $0.050 per 1M input tokens and $0.220 per 1M output tokens, $0.093 blended (three input tokens per output token). The list price last changed on 24 Oct 2026.
- When was gpt-oss-20b released?
- OpenAI released gpt-oss-20b on 5 Aug 2025. BenchLeader has results for it at 4 reasoning-effort settings.
Data as of 24 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “gpt-oss-20b: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/gpt-oss-20b.