Up to four models
Claude Fable 5.1 vs Claude Haiku 5.5
On the index (GPT-4 = 100), Claude Fable 5.1 scores 192 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 Fable 5.1
192175–208
WeightsClosed
ReleasedSep 2026
Price in / out$10 / $50
Context1M
Core tests taken6 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 Fable 5.1 leads Claude Haiku 5.5 by 18 points, ahead on 6 of 6 shared core tests, at 100× the price.
Test by testthird-party results
Claude Fable 5.1Claude Haiku 5.5
Every result as a table
| Test | Claude Fable 5.1 | Claude Haiku 5.5 |
|---|---|---|
| Reasoning & knowledge | ||
| Humanity's Last Exam | 46.5% | 30.1% |
| LiveBench Data Analysis | 80.3% | 72.8% |
| LiveBench IF | 73.0% | 66.5% |
| LiveBench Language | 89.5% | 63.5% |
| LiveBench Reasoning | 91.7% | 84.6% |
| EBR-bench | 57.1% | 43.8% |
| SimpleQA Verified | 70.8% | 23.8% |
| Maths | ||
| FrontierMath (tiers 1-3) | 90.2% | 75.1% |
| AIME-style maths (OTIS mock) | 100.0% | 98.9% |
| LiveBench Mathematics | 97.0% | 92.8% |
| ProofBench v1.1 | 100.0% | 89.0% |
| Chess puzzles | 48.0% | 55.0% |
| FrontierMath (tier 4) | 87.8% | 46.3% |
| Coding | ||
| Terminal-Bench 4.0 | 58.1% | 36.4% |
| LiveBench Agentic Coding | 66.1% | 57.2% |
| LiveBench Coding | 86.4% | 77.9% |
| Code Migration | 54.6% | 44.6% |
| CyberBench v1.1 | 70.4% | 75.5% |
| IOI | 90.8% | 47.3% |
| ProgramBench | 7.0% | 2.5% |
| Vibe Code Bench 1-100 | 28.0% | 20.4% |
| Vibe Code Bench v1.1 | 90.3% | 90.4% |
| FrontierSWE | 56.3% | 43.8% |
| Agents | ||
| APEX-Agents | 68.6% | 56.1% |
| SRE Bench | 22.9% | 5.7% |
| Terminal-Bench-Science | 40.0% | 10.0% |
| DTBench | 97.6% | 95.0% |
| Furniture assembly | 70.0% | 47.5% |
| LMCA | 65.5% | 51.1% |
| Mystery game puzzles | 58.0% | 30.0% |
| Professional work | ||
| Finance Agent (v2) | 58.9% | 54.1% |
| EMB | 76.7% | 59.0% |
| Legal Research Bench | 55.3% | 43.8% |
| MedCode | 53.5% | 51.5% |
| MedScribe | 91.3% | 89.5% |
| Public Benefits Bench v1.1 | 74.9% | 63.2% |
| SAGE | 48.5% | 45.8% |
| Tax Agent Bench | 49.2% | 31.9% |
| Novel problems | ||
| ARC-AGI-2 | 90.0% | 61.3% |
| ARC-AGI-1 | 97.5% | 89.0% |
| MysteryMechanism | 47.8% | 35.1% |
Tests at least two of these models have taken. Bold rows are the core tests.
