OpenAIo3
o3 is an OpenAI proprietary reasoning model, released 16 Apr 2025. Its best configuration ranks #130 of 785 on the BenchLeader Index at 59.5 ±3.5, in the upper half. The ± is the point: 216 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 maths (71) and lowest in reasoning (52). At $3.50 per million tokens blended it is among the most expensive ranked models. Output speed of 134 tokens per second puts it faster than most, with a first answer in 7.4 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 22 Oct 2026.
- Blended price
- $3.50/M
- $2.00 in · $8.00 out
- Output speed
- 134 tok/s
- measured by Artificial Analysis
- First answer
- 7.39 s
- first token 2.48 s
- Context
- 200k
- Full answer
- 11 s
- median, reasoning included
- Cost per run
- –
- Released
- 16 Apr 2025
- Overall index60
- Reasoning52
- Coding62
- Agents & tools70
- Maths71
- Knowledge55
- Instruction following65
- Human preference61
- Multimodal57
- Long context62
- Composite52
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 | 46.4 | #489 | 134 tok/s | 7.39 s | $0.0032 | Maths 41 · Reasoning 48 |
| medium | 53.3 | #290 | 134 tok/s | 7.39 s | $0.0032 | Agents & tools 47 · Coding 37 · Instruction following 54 · Knowledge 58 · Long context 80 · Maths 48 · Multimodal 63 · Reasoning 47 |
| high | 53.9 | #265 | 134 tok/s | 7.39 s | $0.0032 | Coding 53 · Instruction following 49 · Knowledge 58 · Maths 54 · Multimodal 64 · Reasoning 49 |
| not statedbest | 59.5 | #130 | 134 tok/s | 7.39 s | $0.0032 | Agents & tools 70 · Coding 62 · Composite 52 · Human preference 61 · Instruction following 65 · Knowledge 55 · Long context 62 · Maths 71 · Multimodal 57 · Reasoning 52 |
How its index has moved
8 Sept 2026 to 21 Oct 2026- not stated
- high
- medium
- 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 | 79.8%#126 | 80.8%#122 | 81.8%#117 | – | Epoch AI Benchmarking Hub | |
| Humanity's Last Exam | – | 19.2%#23 | 20.3%#21 | – | Scale AI / CAIS | |
| SimpleBench | – | – | 53.1%#43 | – | SimpleBench | |
| Arena Hard Promptseffort not stated | – | – | – | 1442#121 | Arena (LMArena) | |
| GPQA Diamond (AA)not in indexeffort not stated | – | – | – | 82.7%#168 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in indexeffort not stated | – | – | – | 20.1%#215 | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | – | – | 84.1%#52 | – | Vals AI | |
| Kagi LLM Benchmarkeffort not stated | – | – | – | 67.6%#32 | Kagi LLM Benchmark | |
| GPQA Diamond (HELM)not in indexeffort not stated | – | – | – | 75.3%#4 | HELM Capabilities | |
| ARC-AGI-1 | 75.7%#132 | 53.8%#175 | 60.8%#160 | – | ARC Prize | |
| ARC-AGI-2 | 4.0%#185 | 3.0%#193 | 6.5%#168 | – | ARC Prize | |
| CritPteffort not stated | – | – | – | 1.1%#223 | Artificial Analysis | |
| SciPredictnot in indexeffort not stated | – | – | – | 17.9%#10 | Scale AI SEAL | |
| Chess Puzzlesnot in index | 36.0%#49 | 37.0%#44 | 36.0%#49 | – | Epoch AI Benchmarking Hub | |
| Mystery Game Puzzlesnot in index | – | 23.0%#51 | 29.0%#36 | – | Epoch AI Benchmarking Hub | |
| CL-benchnot in index | – | – | 17.8%#14 | – | Epoch AI Benchmarking Hub | |
| LMCAnot in index | – | – | 39.7%#108 | – | Epoch AI Benchmarking Hub | |
| DTBenchnot in index | – | – | 84.8%#106 | – | Epoch AI Benchmarking Hub | |
| ForecastBenchnot in indexeffort not stated | – | – | – | 64.6%#23 | ForecastBench |
Coding
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| SWE-bench Verified (Epoch) | 62.3%#28 | – | – | Epoch AI Benchmarking Hub | |
| WeirdML | – | 52.4%#65 | – | WeirdML | |
| GSO-Bench | – | 8.8%#22 | – | GSO-Bench | |
| Arena Codingeffort not stated | – | – | 1460#129 | Arena (LMArena) | |
| LiveCodeBench | – | 83.9%#43 | – | Vals AI | |
| Aider Polygloteffort not stated | – | – | 81.3%#4 | Aider polyglot leaderboard | |
| SWE-bench Verified (bash only)effort not stated | – | – | 58.4%#25 | SWE-bench | |
| SWE-bench Verified (any scaffold)not in indexeffort not stated | – | – | 58.4%#32 | SWE-bench | |
| CadEvalnot in index | 74.0%#1 | – | – | Epoch AI Benchmarking Hub | |
| ALE-Benchnot in index | – | 933.5#61 | – | Epoch AI Benchmarking Hub |
Agents & tools
| Benchmark | medium | not stated | Source | Trend |
|---|---|---|---|---|
| GDPval | 30.8%#7 | – | OpenAI | |
| Terminal-Bench Hardeffort not stated | – | 37.1%#69 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in indexeffort not stated | – | 80.7%#121 | Artificial Analysis | |
| BFCL Overalleffort not stated | – | 63.0%#7 | Berkeley Function Calling Leaderboard | |
| DeepResearch Benchnot in index | 45.2%#30 | – | DeepResearch Bench | |
| METR Time Horizonsnot in index | 65.4%#15 | 63.6%#18 | METR | |
| BrowseComp-Plusnot in indexeffort not stated | – | 50.5%#2 | BrowseComp-Plus |
Maths
| Benchmark | low | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| FrontierMath Tiers 1–3 | 19.3%#90 | 29.8%#74 | 33.3%#71 | – | Epoch AI Benchmarking Hub | |
| OTIS Mock AIME | 60.0%#163 | 84.4%#98 | 83.9%#105 | – | Epoch AI Benchmarking Hub | |
| MATH Level 5 | – | – | 97.8%#5 | – | Epoch AI Benchmarking Hub | |
| AIME (Vals) | – | – | 85.3%#37 | – | Vals AI | |
| MGSM | – | – | 91.8%#25 | – | Vals AI | |
| IMO 2025 | – | – | 16.7%#4 | – | MathArena | |
| Omni-MATH (HELM)effort not stated | – | – | – | 71.4%#4 | HELM Capabilities | |
| MATH 500not in index | – | – | 94.6%#8 | – | Vals AI | |
| Arena Mathseffort not stated | – | – | – | 1448#80 | Arena (LMArena) |
Knowledge
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| SimpleQA Verified | – | 49.4%#28 | – | Epoch AI Benchmarking Hub | |
| AA-Omniscienceeffort not stated | – | – | -15.6#223 | Artificial Analysis | |
| MMLU-Pro | – | 85.6%#54 | – | Vals AI | |
| LegalBench | – | 83.8%#48 | – | Vals AI | |
| CorpFin | – | 59.7%#72 | – | Vals AI | |
| TaxEval | – | 74.6%#33 | – | Vals AI | |
| MedQA | – | 96.1%#6 | – | Vals AI | |
| PRBench Financeeffort not stated | – | – | 47.7%#15 | Scale AI SEAL | |
| PRBench Legaleffort not stated | – | – | 48.6%#16 | Scale AI SEAL | |
| MultiNRC | 44.5%#16 | 45.5%#14 | – | Scale AI SEAL | |
| MMLU-Pro (HELM)not in indexeffort not stated | – | – | 85.9%#6 | HELM Capabilities | |
| MedCodenot in index | – | 47.3%#34 | – | Vals AI | |
| MedScribenot in index | – | 76.7%#72 | – | Vals AI | |
| AA-Omniscience: accuracynot in indexeffort not stated | – | – | 38.6%#133 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in indexeffort not stated | – | – | 11.9%#385 | Artificial Analysis |
Instruction following
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| IFBencheffort not stated | – | – | 71.4%#56 | Artificial Analysis | |
| MultiChallenge | – | 56.6%#20 | – | Scale AI SEAL | |
| TutorBench | 52.8%#13 | 52.1%#14 | – | Scale AI SEAL | |
| IFEval (HELM)effort not stated | – | – | 86.9%#17 | HELM Capabilities | |
| WildBench (HELM)effort not stated | – | – | 86.1%#4 | HELM Capabilities | |
| Arena Instruction Followingnot in indexeffort not stated | – | – | 1403#134 | Arena (LMArena) |
Human preference
| Benchmark | medium | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Texteffort not stated | – | 1432#108 | Arena (LMArena) | |
| Arena Creative Writingnot in indexeffort not stated | – | 1383#131 | Arena (LMArena) | |
| Arena Multi-turnnot in indexeffort not stated | – | 1420#127 | Arena (LMArena) | |
| Arena Longer Queriesnot in indexeffort not stated | – | 1411#145 | Arena (LMArena) | |
| Lech Mazur Creative Writingnot in index | 83.9%#5 | – | Lech Mazur |
Multimodal
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| Arena Visioneffort not stated | – | – | 1214#83 | Arena (LMArena) | |
| MMMU-Proeffort not stated | – | – | 70.1%#148 | Artificial Analysis | |
| VISTA | 49.6%#12 | 50.1%#10 | – | Scale AI SEAL | |
| MMMU (validation)effort not stated | – | – | 82.9%#3 | MMMU | |
| MMMU-Pro (official)not in indexeffort not stated | – | – | 76.4%#6 | MMMU | |
| MMMU-Pro (Vals)not in index | – | 80.4%#44 | – | Vals AI | |
| MortgageTaxnot in index | – | 65.7%#39 | – | Vals AI | |
| SAGEnot in index | – | 41.8%#56 | – | Vals AI | |
| VTBnot in indexeffort not stated | – | – | 13.7%#8 | Scale AI SEAL | |
| GeoBenchnot in index | 74.0%#10 | 60.0%#19 | – | GeoBench | |
| VPCTnot in index | 52.0%#10 | – | – | Epoch AI Benchmarking Hub |
Long context
| Benchmark | medium | not stated | Source | Trend |
|---|---|---|---|---|
| Fiction.LiveBench 120k | 100.0%#1 | – | Fiction.live | |
| AA-LCReffort not stated | – | 74.7%#159 | Artificial Analysis |
Composite
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | 146.7#73 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 20.2#233 | Artificial Analysis | |
| HELM Capabilities meannot in index | 81.1%#3 | HELM Capabilities |
Safety & honesty
| Benchmark | medium | high | not stated | Source | Trend |
|---|---|---|---|---|---|
| FORTRESSnot in indexeffort not stated | – | – | 16.0%#52 | Scale AI SEAL | |
| MASKnot in indexpossible contamination | 82.6%#20 | 84.5%#19 | – | Scale AI SEAL | |
| PropensityBenchnot in indexeffort not stated | – | – | 10.5%#14 | Scale AI SEAL |
Where it wins
Benchmarks where this configuration ranks in the top five of every configuration measured.
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 |
|---|---|---|---|---|---|---|
| OpenAI | 53 tok/s | 1.91 s | $2.00 | $8.00 | 200k | – |
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 $0.500 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0032 | $0.0028 | 9.6 s |
| Summarize a 30-page report | 12,000 / 600 | $0.029 | $0.015 | 11.9 s |
| Code edit | 6,000 / 1,500 | $0.024 | $0.017 | 18.6 s |
| Agentic coding session | 60,000 / 4,000 | $0.152 | $0.085 | 37.2 s |
| Structured extraction | 2,000 / 200 | $0.0056 | $0.0034 | 8.9 s |
Compare o3 with…
See also
Questions people ask
- What are o3's benchmark scores?
- As of 25 Oct 2026, o3 scores 59.5 ±3.5 on the BenchLeader Index, strongest in maths (71) and weakest in reasoning (52). That rests on 22 independent benchmark results, and every score on this page links to its publisher.
- How does o3 rank?
- o3 is #130 of 785 ranked configurations on 25 Oct 2026. 216 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 o3 cost per 1M tokens?
- As of 25 Oct 2026, o3 costs $2.00 per 1M input tokens and $8.00 per 1M output tokens, $3.50 blended (three input tokens per output token).
- Is o3 better than GPT-5.6 Sol?
- As of 25 Oct 2026, GPT-5.6 Sol (max) leads on quality (68.0 vs 59.5 on the BenchLeader Index), but o3 costs 2.3× less, so it is the better buy unless you need the extra quality. The head-to-head page has the benchmark-by-benchmark detail.
- When was o3 released?
- OpenAI released o3 on 16 Apr 2025. BenchLeader has results for it at 4 reasoning-effort settings.
Data as of 25 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “o3: benchmarks, pricing, speed and rank”, data as of 25 Oct 2026, https://www.benchleader.com/models/o3.