Up to four models
Claude Sonnet 5.5 vs Claude Opus 5
On the index (GPT-4 = 100), Claude Sonnet 5.5 scores 187 and Claude Opus 5 185. 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 5.5
187171–203
WeightsClosed
ReleasedSep 2026
Price in / out$2.0 / $10
Context1M
Core tests taken7 of 9
Claude Opus 5
185169–201
WeightsClosed
ReleasedJul 2026
Price in / out$5.0 / $25
Context1M
Core tests taken6 of 9
Claude Sonnet 5.5 and Claude Opus 5 are too close to call (2 points apart, within the uncertainty), ahead on 4 of 6 shared core tests, and costs 2.5× less.
Test by testthird-party results
Claude Sonnet 5.5Claude Opus 5
Every result as a table
| Test | Claude Sonnet 5.5 | Claude Opus 5 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 95.6% | 93.9% |
| LiveBench Data Analysis | 78.6% | 74.6% |
| LiveBench IF | 70.5% | 63.8% |
| LiveBench Language | 83.4% | 88.7% |
| LiveBench Reasoning | 91.6% | 91.2% |
| SimpleBench | 75.9% | 80.6% |
| SimpleQA Verified | 46.5% | 59.9% |
| Maths | ||
| FrontierMath (tiers 1-3) | 88.8% | 85.6% |
| AIME-style maths (OTIS mock) | 100.0% | 98.9% |
| LiveBench Mathematics | 96.7% | 95.7% |
| ProofBench v1.1 | 100.0% | 99.0% |
| Chess puzzles | 59.0% | 43.0% |
| FrontierMath (tier 4) | 80.5% | 73.2% |
| Coding | ||
| Terminal-Bench 4.0 | 64.1% | 53.5% |
| LiveBench Agentic Coding | 56.3% | 65.2% |
| LiveBench Coding | 91.4% | 81.5% |
| Code Migration | 69.8% | 57.5% |
| CyberBench v1.1 | 59.6% | 65.4% |
| IOI | 83.1% | 84.3% |
| ProgramBench | 6.5% | 3.0% |
| Vibe Code Bench v1.1 | 92.4% | 88.4% |
| FrontierSWE | 61.9% | 52.0% |
| MirrorCode | 97.8% | 87.8% |
| Agents | ||
| APEX-Agents | 75.5% | 65.8% |
| BioMysteryBench | 81.1% | 79.3% |
| SRE Bench | 30.1% | 12.2% |
| Terminal-Bench-Science | 47.1% | 27.1% |
| DTBench | 97.3% | 97.9% |
| Furniture assembly | 75.0% | 60.8% |
| LMCA | 68.3% | 64.5% |
| Mystery game puzzles | 65.0% | 59.0% |
| Professional work | ||
| Finance Agent (v2) | 58.1% | 58.6% |
| EMB | 75.7% | 73.6% |
| Legal Research Bench | 48.1% | 55.3% |
| MedCode | 52.9% | 63.6% |
| MedScribe | 91.1% | 91.0% |
| Public Benefits Bench v1.1 | 67.2% | 76.9% |
| Tax Agent Bench | 42.0% | 47.1% |
| Novel problems | ||
| ARC-AGI-2 | 83.9% | 90.4% |
| ARC-AGI-1 | 96.5% | 97.5% |
| ARC-AGI-3 | 24.3% | 30.2% |
| MysteryMechanism | 49.1% | 37.4% |
Tests at least two of these models have taken. Bold rows are the core tests.