Claude Opus 4.5 (Thinking): benchmark results
Claude Opus 4.5 evaluated with thinking enabled. Provider: Anthropic. Released 2025-11-24. Access: API.
Aggregate Rating 1740 ± 7, rank #259 of 2168 rated models, from 336 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA Omniscience - Software Engineering (SWE) - Go | 70 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - Julia | 68 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - PHP | 84 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - Python | 74.5 | Accuracy (%) | 100 |
| BaxBench - No Security Reminder - Correct | 86.22 | Correct solutions, pass@1 (%) | 100 |
| BaxBench - No Security Reminder - Correct & Secure | 56.12 | Correct and secure solutions, sec_pass@1 (%) | 100 |
| EvoCodeBench (LeetCode) - Kotlin | 83 | Pass rate (%): problems of 100 whose Kotlin submission is ac | 100 |
| FormalProofBench | 33.5 | Accuracy (%): share of FormalProofBench's 200 private gradua | 100 |
| MCP-Atlas (Opus 4.6 System Card) | 62.3 | Score (%) | 100 |
| MM-JudgeBias - Image Dominance | 0.59 | Bias-Deviation (0 to 1) on MM-JudgeBias (about 200 image-que | 100 |
| OpenCompass Agent - Multi-Turn (CompassBench 2601) | 84.8 | Score (%) | 100 |
| OpenCompass LLM - Agent (CompassBench 2601) | 76.2 | Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=claude-opus-4-5-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.