Up to four models
Claude Sonnet 5.5 vs Claude Haiku 5.5
On the index (GPT-4 = 100), Claude Sonnet 5.5 scores 191 and Claude Haiku 5.5 174. 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
191175–207
WeightsClosed
ReleasedSep 2026
Price in / out$2.0 / $10
Context1M
Core tests taken7 of 9
Claude Haiku 5.5
174158–191
WeightsClosed
ReleasedOct 2026
Price in / out$0.10 / $0.50
Context1M
Core tests taken7 of 9
Claude Sonnet 5.5 leads Claude Haiku 5.5 by 17 points, ahead on 7 of 7 shared core tests, at 20× the price.
Test by testthird-party results
Claude Sonnet 5.5Claude Haiku 5.5
Every result as a table
| Test | Claude Sonnet 5.5 | Claude Haiku 5.5 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 95.6% | 89.6% |
| Humanity's Last Exam | 46.1% | 30.1% |
| LiveBench Data Analysis | 78.6% | 72.8% |
| LiveBench IF | 70.5% | 66.5% |
| LiveBench Language | 83.4% | 63.5% |
| LiveBench Reasoning | 91.6% | 84.6% |
| SimpleQA Verified | 46.5% | 23.8% |
| Maths | ||
| FrontierMath (tiers 1-3) | 88.8% | 75.1% |
| AIME-style maths (OTIS mock) | 100.0% | 98.9% |
| LiveBench Mathematics | 96.7% | 92.8% |
| ProofBench v1.1 | 100.0% | 89.0% |
| Chess puzzles | 59.0% | 55.0% |
| FrontierMath (tier 4) | 80.5% | 46.3% |
| Coding | ||
| Terminal-Bench 4.0 | 64.1% | 36.4% |
| LiveBench Agentic Coding | 56.3% | 57.2% |
| LiveBench Coding | 91.4% | 77.9% |
| Code Migration | 69.8% | 44.6% |
| CyberBench v1.1 | 59.6% | 75.5% |
| IOI | 83.1% | 47.3% |
| ProgramBench | 6.5% | 2.5% |
| Vibe Code Bench v1.1 | 92.4% | 90.4% |
| FrontierSWE | 61.9% | 43.8% |
| Agents | ||
| APEX-Agents | 75.5% | 56.1% |
| BioMysteryBench | 81.1% | 70.4% |
| SRE Bench | 30.1% | 5.7% |
| Terminal-Bench-Science | 47.1% | 10.0% |
| DTBench | 97.3% | 95.0% |
| Furniture assembly | 75.0% | 47.5% |
| LMCA | 68.3% | 51.1% |
| Mystery game puzzles | 65.0% | 30.0% |
| Professional work | ||
| Finance Agent (v2) | 58.1% | 54.1% |
| EMB | 75.7% | 59.0% |
| HLAB-Vals-1.1 | 10.4% | 4.2% |
| Legal Research Bench | 48.1% | 43.8% |
| MedCode | 52.9% | 51.5% |
| MedScribe | 91.1% | 89.5% |
| Public Benefits Bench v1.1 | 67.2% | 63.2% |
| SAGE | 51.8% | 45.8% |
| Tax Agent Bench | 42.0% | 31.9% |
| Novel problems | ||
| ARC-AGI-2 | 83.9% | 61.3% |
| ARC-AGI-1 | 96.5% | 89.0% |
| ARC-AGI-3 | 24.3% | 0.5% |
| MysteryMechanism | 49.1% | 35.1% |
Tests at least two of these models have taken. Bold rows are the core tests.
