Up to four models
Claude Haiku 5.5 vs Grok 4.6
On the index (GPT-4 = 100), Claude Haiku 5.5 scores 168 and Grok 4.6 167. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Claude Haiku 5.5
168152–184
WeightsClosed
ReleasedOct 2026
Price in / out$0.10 / $0.50
Context1M
Core tests taken7 of 9
Grok 4.6
167151–183
WeightsClosed
ReleasedAug 2026
Price in / out$2.0 / $6.0
Context500K
Core tests taken6 of 9
Claude Haiku 5.5 and Grok 4.6 are too close to call (1 point apart, within the uncertainty), ahead on 3 of 6 shared core tests, and costs 15× less.
Test by testthird-party results
Claude Haiku 5.5Grok 4.6
Every result as a table
| Test | Claude Haiku 5.5 | Grok 4.6 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 89.6% | 94.0% |
| LiveBench Data Analysis | 72.8% | 73.9% |
| LiveBench IF | 66.5% | 71.9% |
| LiveBench Language | 63.5% | 83.7% |
| LiveBench Reasoning | 84.6% | 90.5% |
| EBR-bench | 43.8% | 30.5% |
| SimpleBench | 59.2% | 75.9% |
| SimpleQA Verified | 23.8% | 49.3% |
| Maths | ||
| FrontierMath (tiers 1-3) | 75.1% | 66.0% |
| AIME-style maths (OTIS mock) | 98.9% | 99.2% |
| LiveBench Mathematics | 92.8% | 92.6% |
| ProofBench v1.1 | 89.0% | 51.0% |
| Chess puzzles | 55.0% | 33.0% |
| FrontierMath (tier 4) | 46.3% | 31.7% |
| Coding | ||
| Terminal-Bench 4.0 | 36.4% | 17.2% |
| LiveBench Agentic Coding | 57.2% | 57.0% |
| LiveBench Coding | 77.9% | 76.8% |
| Code Migration | 44.6% | 44.6% |
| CyberBench v1.1 | 75.5% | 67.7% |
| IOI | 47.3% | 47.6% |
| ProgramBench | 2.5% | 1.0% |
| Vibe Code Bench 1-100 | 20.4% | 14.8% |
| Vibe Code Bench v1.1 | 90.4% | 76.2% |
| FrontierSWE | 43.8% | 25.3% |
| MirrorCode | 78.9% | 41.1% |
| Agents | ||
| APEX-Agents | 56.1% | 65.3% |
| BioMysteryBench | 70.4% | 72.2% |
| Terminal-Bench-Science | 10.0% | 11.4% |
| DTBench | 95.0% | 97.3% |
| Furniture assembly | 47.5% | 40.0% |
| LMCA | 51.1% | 48.4% |
| Mystery game puzzles | 30.0% | 34.0% |
| Professional work | ||
| Finance Agent (v2) | 54.1% | 53.7% |
| EMB | 59.0% | 62.7% |
| Legal Research Bench | 43.8% | 48.1% |
| MedCode | 51.5% | 44.7% |
| MedScribe | 89.5% | 86.5% |
| Public Benefits Bench v1.1 | 63.2% | 66.8% |
| Tax Agent Bench | 31.9% | 33.7% |
| Novel problems | ||
| ARC-AGI-2 | 61.3% | 67.1% |
| ARC-AGI-1 | 89.0% | 87.5% |
| ARC-AGI-3 | 0.5% | 2.1% |
| MysteryMechanism | 35.1% | 30.6% |
Tests at least two of these models have taken. Bold rows are the core tests.