Up to four models
Claude Opus 4.5 vs Claude Opus 4.1
On the index (GPT-4 = 100), Claude Opus 4.5 scores 152 and Claude Opus 4.1 140. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Claude Opus 4.5
152136–168
WeightsClosed
ReleasedNov 2025
Price in / out$5.0 / $25
Context200K
Core tests taken5 of 9
Claude Opus 4.1
140124–156
WeightsClosed
ReleasedAug 2025
Price in / out$15 / $75
Context200K
Core tests taken5 of 9
Claude Opus 4.5 leads Claude Opus 4.1 by 12 points, ahead on 5 of 5 shared core tests, and costs 3.0× less.
Test by testthird-party results
Claude Opus 4.5Claude Opus 4.1
Every result as a table
| Test | Claude Opus 4.5 | Claude Opus 4.1 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 86.1% | 77.3% |
| Humanity's Last Exam | 25.2% | 11.5% |
| GPQA Diamond (Vals AI) | 85.9% | 76.3% |
| MMLU Pro | 87.3% | 87.9% |
| MMMU Pro | 83.0% | 77.5% |
| EBR-bench | 14.3% | 7.9% |
| SimpleBench | 62.0% | 60.0% |
| VPCT | 40.0% | 35.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 34.4% | 12.6% |
| AIME-style maths (OTIS mock) | 86.1% | 68.9% |
| MGSM | 95.2% | 94.4% |
| FrontierMath (2025 set) | 20.7% | 7.2% |
| FrontierMath tier 4 (2025 set) | 4.2% | 4.2% |
| FrontierMath (tier 4) | 4.9% | 2.4% |
| Coding | ||
| SWE-bench Verified | 76.6% | 73.4% |
| Terminal-Bench 2.0 (Terminus 2 runs) | 57.8% | 38.0% |
| LiveCodeBench | 83.7% | 66.5% |
| WeirdML | 63.7% | 45.9% |
| Agents | ||
| DTBench | 89.9% | 80.0% |
| DeepResearch Bench | 54.8% | 48.3% |
| LMCA | 44.5% | 37.1% |
| Mystery game puzzles | 22.0% | 21.0% |
| Professional work | ||
| LegalBench | 84.6% | 83.5% |
| MedCode | 49.2% | 47.2% |
| MedScribe | 85.3% | 73.9% |
| MortgageTax | 68.7% | 61.1% |
| TaxEval v2 | 74.9% | 73.7% |
Task length · METR45 sec40 sec
Tests at least two of these models have taken. Bold rows are the core tests.