Up to four models
Claude Sonnet 4.5 vs Claude Sonnet 4
On the index (GPT-4 = 100), Claude Sonnet 4.5 scores 144 and Claude Sonnet 4 133. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Claude Sonnet 4.5
144129–160
WeightsClosed
ReleasedSep 2025
Price in / out$3.0 / $15
Context1M
Core tests taken6 of 9
Claude Sonnet 4.5 leads Claude Sonnet 4 by 11 points, ahead on 4 of 4 shared core tests.
Test by testthird-party results
Claude Sonnet 4.5Claude Sonnet 4
Every result as a table
| Test | Claude Sonnet 4.5 | Claude Sonnet 4 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 82.3% | 79.2% |
| Humanity's Last Exam | 13.7% | 7.8% |
| GPQA Diamond (Vals AI) | 81.6% | 75.0% |
| MMLU Pro | 87.4% | 83.9% |
| MMMU Pro | 79.3% | 74.9% |
| GeoBench | 60.0% | 37.0% |
| SimpleBench | 54.3% | 45.5% |
| VPCT | 39.8% | 34.0% |
| Maths | ||
| AIME-style maths (OTIS mock) | 77.8% | 71.1% |
| MATH (level 5) | 97.7% | 84.4% |
| MGSM | 94.3% | 93.0% |
| FrontierMath (2025 set) | 15.2% | 4.1% |
| FrontierMath tier 4 (2025 set) | 4.2% | 0.0% |
| Coding | ||
| SWE-bench Pro (public set) | 43.6% | 42.7% |
| LiveCodeBench | 73.0% | 62.4% |
| GSO-Bench | 14.7% | 4.9% |
| WeirdML | 47.7% | 46.1% |
| Agents | ||
| DTBench | 83.2% | 77.1% |
| DeepResearch Bench | 52.6% | 46.6% |
| LMCA | 38.8% | 29.0% |
| Professional work | ||
| CorpFin v2 | 62.0% | 61.2% |
| LegalBench | 84.1% | 83.0% |
| MedCode | 44.1% | 35.0% |
| MedScribe | 84.5% | 72.4% |
| MortgageTax | 64.0% | 62.5% |
| TaxEval v2 | 73.3% | 72.0% |
| Novel problems | ||
| ARC-AGI-2 | 13.6% | 5.9% |
| ARC-AGI-1 | 63.7% | 40.0% |
Task length · METR40 sec37 sec
Tests at least two of these models have taken. Bold rows are the core tests.