Claude Opus 5
AnthropicProprietaryReasoningVerifiedReleased Jul 24, 2026Input: text, image, file
Overall score
80.1
Rank 8 of 222 · 77.2–83.0
Context window
1M
1,000,000 tokens
Price per 1M tokens
$5.00 in · $25.00 out
✓ Verified on Anthropic's pricing page
Blended price
$10.00
3:1 input:output tokens
Output speed
—
Not measured yet
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Knowledge#795.3
- 2Multimodal#292.2
- 3Math91.6
- 4Coding#990.4
- 5Agentic & tool use#485.2
- 6Reasoning#881.0
Benchmark results (76)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Agentic & tool use
| BrowseComp | 90.8 | #6 of 50 |
| MCP Atlas | 85.8 | #3 of 42 |
| MazeBench | 12 | #5 of 42 |
| HLE w/ tools | 64.7 | #2 of 27 |
| AutomationBench | 26.0 | #24 of 25 |
| OSWorld 2.0 | 70.6 | #2 of 24 |
| Toolathlon-Verified | 80.6 | #1 of 22 |
| DeepSearchQA | 95.0 | #2 of 20 |
| ApprenticeBench | 36 | #3 of 18 |
| Terminal-Bench 3.0 | 42.7 | #1 of 13 |
| DRACO | 88.6 | #1 of 6 |
| Toolathlon Verified Pass@3 | 87.0 | #1 of 4 |
| Toolathlon Verified Pass³ | 73.1 | #1 of 4 |
| Toolathlon Verified avg. turns(lower is better) | 23.5 | #1 of 4 |
| LAB all-pass (Harvey held-out) | 11.7 | #1 of 3 |
| LAB criterion-pass (Harvey held-out) | 94.1 | #1 of 3 |
| MCP-Atlas claim coverage | 89.1 | #1 of 2 |
| BrowseComp (10-agent, prerelease) | 93.6 | #1 of 1 |
| LAB all-pass (Anthropic harness) | 23.58 | #1 of 1 |
| LAB criterion-pass (Anthropic harness) | 93.74 | #1 of 1 |
Knowledge
| HLE | 64.7 | #2 of 69 |
| HLE w/o tools | 56.3 | #5 of 43 |
| HealthBench Professional | 59.8 | #11 of 16 |
| LABBench2 | 84.2 | #3 of 7 |
| HealthBench (raw) | 67.1 | #3 of 7 |
| HealthBench (length-adjusted) | 57.8 | #5 of 7 |
| HealthBench Professional (raw) | 73.4 | #3 of 7 |
| HLE-Verified | 54.4 | #3 of 6 |
| BioMysteryBench (human-difficult) | 49.4 | #3 of 6 |
| BioMysteryBench (human-solvable) | 90.1 | #1 of 5 |
| SpatialBench Verified | 72.5 | #1 of 3 |
| SingleCellBench | 60.6 | #2 of 3 |
| Protein Design | 42.5 | #3 of 3 |
| Protocols (troubleshooting) | 61.1 | #3 of 3 |
| Protocols (understanding) | 78.4 | #1 of 3 |
| ProteinGym Hard | 47.7 | #1 of 1 |
| Organic chemistry V2 | 61.6 | #1 of 1 |
Coding
| SWE-bench Verified | 96.0 | #1 of 98 |
| SWE-bench Pro | 79.2 | #6 of 69 |
| DeepSWE | 68.8 | #20 of 50 |
| Bug Hunt Bench | 27.0 | #13 of 23 |
| FrontierSWE v2 | 52.0 | #6 of 20 |
| FrontierCode 1.1 Main | 53.4 | #3 of 16 |
| ProgramBench | 93.0 | #1 of 15 |
| PostTrainBench v1.1 | 35.0 | #7 of 15 |
| VulcanBench v3 | 87.0 | #4 of 14 |
| FrontierCode 1.1 Extended | 63.6 | #2 of 8 |
| VulcanBench CII v1 | 96.4 | #1 of 3 |
| ProgramBench (episode 1) | 83.0 | #1 of 1 |
Multimodal
| Design Arena Website | 1309 | #11 of 91 |
| OfficeQA Pro | 66.9 | #2 of 13 |
| SWE-bench Multimodal | 59.4 | #2 of 7 |
| Chartography (no tools) | 29.6 | #5 of 5 |
| Chartography (tools) | 83.0 | #3 of 5 |
| BenchCAD Vision2Code (tools) | 0.821 | #4 of 4 |
| BenchCAD Vision2Code (no tools) | 0.366 | #3 of 3 |
| OfficeQA | 78.1 | #2 of 3 |
| GDP.pdf (no tools) | 83.4 | #1 of 2 |
| GDP.pdf (tools) | 85.5 | #1 of 1 |
Math
| FrontierMath v2 (Tiers 1-3) | 85.614 | #11 of 64 |
| FrontierMath v2 (Tier 4) | 73.171 | #11 of 51 |
| IMO 2026 | 42 | #1 of 2 |
| RiemannBench (no tools) | 60.0 | #1 of 1 |
| RiemannBench (tools) | 79.0 | #1 of 1 |
| ArXivMath Jun. 2026 (no tools) | 90.8 | #1 of 1 |
| ArXivMath Jun. 2026 (tools) | 91.3 | #1 of 1 |
Multilingual
| SWE Multilingual | 89.5 | #4 of 46 |
| INCLUDE | 89.8 | #1 of 10 |
| GMMLU | 92.5 | #2 of 3 |
| MILU | 92.1 | #2 of 3 |
Reasoning
| ARC-AGI-2 | 90.4 | #6 of 41 |
| ARC-AGI-1 | 97.50 | #4 of 36 |
| ARC-AGI-3 | 30.2 | #4 of 22 |
Instruction following
| Gray Swan IPI (15 attempts)(lower is better) | 4.6 | #4 of 12 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| ✓ Official (Anthropic) checked Oct 17, 2026 · Claude Opus 5 | $5.00 | $25.00 | $0.50 |
| BenchLM list price | $5.00 | $25.00 | $0.50 |
| OpenRouter anthropic/claude-opus-5 | $5.00 | $25.00 | $0.50 |
Anthropic's July 24, 2026 Claude Opus 5 announcement lists the claude-opus-5 API model at $5 input / $25 output per million tokens, the same base price as Opus 4.8. Fast mode runs around 2.5× the default speed and is available at twice the base price.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text(Max) | 1508 | #4 · 32.5K votes |
| WebDev (coding)(Max) | 1689 | #8 · 18.6K votes |
| Vision(High) | 1320 | #3 · 17.7K votes |
| Search(high) | 1242 | #6 · 13.3K votes |
| Documents(high) | 1516 | #1 · 8.8K votes |
Speed
Measured by us through OpenRouter (median of trials).
Not measured yet.
Related models
Same family
- Claude Opus 5.586.7
- Claude Opus 4.870.0
- Claude Sonnet 564.9
Similar score
- Claude Fable 581.3 · $20.00
- GPT-6 Sol78.6 · $4.00
- Claude Sonnet 5.582.2 · $4.00
- Gemini 4 Argon82.9 · $4.00
- Claude Fable 5.183.2 · $20.00