Claude Opus 4.5
Anthropic·
LLMsflagshipcloud
Overview
Next-generation flagship model with breakthrough reasoning and research capabilities.
Capabilities and innovations
Advanced reasoningScientific researchLong-context understandingMulti-step planningEnhanced Constitutional AIExtended Context ProcessingAdvanced Tool Integration
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 93.2% | unsourced |
| Humanity’s Last Exam (no tools) | 44.8% | unsourced |
| SWE-bench Verified | 76.9% | unsourced |
| MMLU | 94.5% | unsourced |
| MMLU-Pro | 87.26% | unsourced |
- SWE-bench Pro: not reported by the vendor
- Terminal-Bench 2.x: benchmark did not exist at release
- MMMU-Pro: benchmark did not exist at release
Reliability
- Hallucination rate (Vectara HHEM)
- 10.9% (lower is better)independent
- Function calling (BFCL)
- 77.47% (higher is better)independent
Links
More from Anthropic
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.