Up to four models
Claude Sonnet 4.6 vs Claude Sonnet 4.5
On the index (GPT-4 = 100), Claude Sonnet 4.6 scores 159 and Claude Sonnet 4.5 144. 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 4.6
159143–175
WeightsClosed
ReleasedFeb 2026
Price in / out$3.0 / $15
Context1M
Core tests taken5 of 9
Claude Sonnet 4.5
144129–160
WeightsClosed
ReleasedSep 2025
Price in / out$3.0 / $15
Context1M
Core tests taken6 of 9
Claude Sonnet 4.6 leads Claude Sonnet 4.5 by 14 points, ahead on 3 of 3 shared core tests, at a similar price.
Test by testthird-party results
Claude Sonnet 4.6Claude Sonnet 4.5
Every result as a table
| Test | Claude Sonnet 4.6 | Claude Sonnet 4.5 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 87.4% | 82.3% |
| GPQA Diamond (Vals AI) | 85.6% | 81.6% |
| MMLU Pro | 87.3% | 87.4% |
| MMMU Pro | 83.6% | 79.3% |
| GeoBench | 73.0% | 60.0% |
| SimpleQA Verified | 35.5% | 30.7% |
| Maths | ||
| AIME-style maths (OTIS mock) | 85.8% | 77.8% |
| Chess puzzles | 23.0% | 10.0% |
| FrontierMath (2025 set) | 32.4% | 15.2% |
| FrontierMath tier 4 (2025 set) | 8.3% | 4.2% |
| Coding | ||
| SWE-bench Verified | 75.2% | 71.3% |
| LiveCodeBench | 82.1% | 73.0% |
| SWE-bench | 77.4% | 70.0% |
| Vibe Code Bench v1.1 | 51.5% | 22.6% |
| WeirdML | 66.1% | 47.7% |
| Agents | ||
| DTBench | 89.9% | 83.2% |
| DeepResearch Bench | 54.9% | 52.6% |
| LMCA | 46.5% | 38.8% |
| Mystery game puzzles | 16.0% | 17.0% |
| Professional work | ||
| CorpFin v2 | 65.3% | 62.0% |
| LegalBench | 82.1% | 84.1% |
| MortgageTax | 67.7% | 64.0% |
| TaxEval v2 | 77.1% | 73.3% |
| Novel problems | ||
| ARC-AGI-2 | 60.4% | 13.6% |
| ARC-AGI-1 | 86.5% | 63.7% |
Task length · METR40 sec
Tests at least two of these models have taken. Bold rows are the core tests.