AnthropicReasoning modelClaude Opus 4
Best configuration ranks #198 of 610 on the BenchLeader Index at 54.4 ±4.2 (thinking reasoning effort). Last measured 2 Sept 2026. Released 22 May 2025.
- Blended price
- $30.00/M
- $15.00 in · $75.00 out
- Output speed
- 20 tok/s
- First answer
- 3.02 s
- Context
- 200k
- Overall index54
- Reasoning49
- Coding54
- Agents & tools61
- Knowledge48
- Instruction following52
- Human preference61
- Multimodal55
- Long context61
- Composite55
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 |
|---|---|---|---|---|---|---|
| thinkingbest | 54.4 | #198 | 20 tok/s | 3.02 s | $0.029 | Agents & tools 61 · Coding 54 · Composite 55 · Human preference 61 · Instruction following 52 · Knowledge 48 · Long context 61 · Multimodal 55 · Reasoning 49 |
| default | 50.7 | #278 | 20 tok/s | 3.02 s | $0.029 | Agents & tools 56 · Coding 53 · Composite 50 · Human preference 60 · Instruction following 42 · Knowledge 55 · Long context 37 · Maths 54 · Multimodal 54 · Reasoning 48 |
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 | thinking | default | Source |
|---|---|---|---|
| GPQA Diamond | – | 76.3%#129 | Epoch AI Benchmarking Hub |
| Humanity's Last Exam | – | 10.7%#26 | Scale AI / CAIS |
| SimpleBench | – | 58.8%#32 | SimpleBench |
| LMArena Hard Prompts | 1456#87 | 1436#112 | LMArena |
| GPQA Diamond (AA)not in index | 79.6%#191 | 70.1%#284 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 12.3%#217 | 6.2%#331 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | – | 71.7%#87 | Vals AI |
| Kagi LLM Benchmark | 74.3%#15 | 59.6%#52 | Kagi LLM Benchmark |
| ARC-AGI-1 | 35.7%#137 | 22.5%#152 | ARC Prize |
| ARC-AGI-2 | 8.6%#109 | 1.3%#153 | ARC Prize |
Coding
| Benchmark | thinking | default | Source |
|---|---|---|---|
| SWE-bench Verified (Epoch) | – | 70.7%#25 | Epoch AI Benchmarking Hub |
| WeirdML | – | 43.7%#84 | WeirdML |
| GSO-Bench | – | 6.9%#19 | GSO-Bench |
| LMArena Coding | 1498#65 | 1465#107 | LMArena |
| LiveCodeBench | 70.2%#88 | 62.6%#99 | Vals AI |
| Aider Polyglot | – | 72.0%#7 | Aider polyglot leaderboard |
| SWE-bench Verified (bash only) | – | 67.6%#15 | SWE-bench |
| SWE-bench Verified (any scaffold)not in index | – | 73.2%#10 | SWE-bench |
Agents & tools
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Cybench | – | 38.0%#7 | Cybench |
| Terminal-Bench Hard | 31.1%#104 | – | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 73.4%#140 | – | Artificial Analysis |
Maths
| Benchmark | thinking | default | Source |
|---|---|---|---|
| OTIS Mock AIME | – | 64.4%#139 | Epoch AI Benchmarking Hub |
| MATH Level 5 | – | 85.0%#26 | Epoch AI Benchmarking Hub |
| AIME (Vals) | – | 41.3%#67 | Vals AI |
| MGSM | – | 93.8%#8 | Vals AI |
Knowledge
| Benchmark | thinking | default | Source |
|---|---|---|---|
| MMLU-Pro | – | 86.2%#46 | Vals AI |
| LegalBench | – | 83.1%#54 | Vals AI |
| TaxEval | – | 71.9%#69 | Vals AI |
| MedQA | – | 92.9%#26 | Vals AI |
| MultiNRC | 33.9%#22 | 29.0%#24 | Scale AI SEAL |
Instruction following
| Benchmark | thinking | default | Source |
|---|---|---|---|
| IFBench | 53.7%#152 | 43.3%#221 | Artificial Analysis |
| MultiChallenge | 58.6%#13 | – | Scale AI SEAL |
| TutorBench | 49.7%#19 | 45.5%#25 | Scale AI SEAL |
Human preference
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Text | 1426#100 | 1414#115 | LMArena |
Multimodal
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Vision | 1192#78 | 1175#86 | LMArena |
| VISTA | 47.0%#18 | 43.5%#32 | Scale AI SEAL |
| MMMU (validation) | – | 76.5%#13 | MMMU |
Long context
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Fiction.LiveBench 120k | – | 37.5%#26 | Fiction.live |
| AA-LCR | 69.3%#170 | – | Artificial Analysis |
Composite
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | 142.7#83 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 20.6#168 | 16.6#209 | 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 |
|---|---|---|---|---|---|---|
| Google Vertex | 20 tok/s | 3.02 s | $15.00 | $75.00 | 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 $1.50 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.029 | $0.025 | 18.0 s |
| Summarise a 30-page report | 12,000 / 600 | $0.225 | $0.103 | 33.0 s |
| Code edit | 6,000 / 1,500 | $0.203 | $0.142 | 1.3 min |
| Agentic coding session | 60,000 / 4,000 | $1.20 | $0.593 | 3.4 min |
| Structured extraction | 2,000 / 200 | $0.045 | $0.025 | 13.0 s |
See also
Data as of 9 Sept 2026. Compare these configurations.