Up to four models
Grok 4.6 vs Grok 4.5
On the index (GPT-4 = 100), Grok 4.6 scores 171 and Grok 4.5 165. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Grok 4.6
171155–188
WeightsClosed
ReleasedAug 2026
Price in / out$2.0 / $6.0
Context500K
Core tests taken6 of 9
Grok 4.5
165149–182
WeightsClosed
ReleasedJul 2026
Price in / out$2.0 / $6.0
Context500K
Core tests taken6 of 9
Grok 4.6 leads Grok 4.5 by 6 points, ahead on 6 of 6 shared core tests, at a similar price.
Test by testthird-party results
Grok 4.6Grok 4.5
Every result as a table
| Test | Grok 4.6 | Grok 4.5 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 94.0% | 93.4% |
| LiveBench Data Analysis | 73.9% | 73.0% |
| LiveBench IF | 71.9% | 71.5% |
| LiveBench Language | 83.7% | 82.8% |
| LiveBench Reasoning | 90.5% | 87.2% |
| GPQA Diamond (Vals AI) | 94.7% | 92.9% |
| MMLU Pro | 89.4% | 89.2% |
| SimpleBench | 75.9% | 70.0% |
| SimpleQA Verified | 49.3% | 48.3% |
| Maths | ||
| FrontierMath (tiers 1-3) | 66.0% | 57.2% |
| AIME-style maths (OTIS mock) | 99.2% | 97.8% |
| LiveBench Mathematics | 92.6% | 90.8% |
| ProofBench v1.1 | 51.0% | 31.0% |
| Chess puzzles | 33.0% | 35.0% |
| FrontierMath (tier 4) | 31.7% | 24.4% |
| Coding | ||
| Terminal-Bench 4.0 | 17.2% | 8.6% |
| LiveBench Agentic Coding | 57.0% | 56.5% |
| LiveBench Coding | 76.8% | 68.6% |
| Code Migration | 44.6% | 36.6% |
| CyberBench v1.1 | 67.7% | 71.0% |
| IOI | 47.6% | 40.6% |
| LiveCodeBench | 88.2% | 87.4% |
| ProgramBench | 1.0% | 0.0% |
| SWE-bench | 95.6% | 86.6% |
| Vibe Code Bench v1.1 | 76.2% | 69.0% |
| DeepSWE | 67.5% | 53.8% |
| WeirdML | 67.3% | 46.4% |
| Agents | ||
| APEX-Agents | 65.3% | 56.2% |
| SkillsBench | 55.8% | 66.0% |
| DTBench | 97.3% | 96.5% |
| Furniture assembly | 40.0% | 22.5% |
| LMCA | 48.4% | 45.2% |
| Professional work | ||
| Finance Agent (v2) | 53.7% | 48.4% |
| CorpFin v2 | 66.2% | 67.4% |
| EMB | 62.7% | 52.9% |
| LegalBench | 86.3% | 86.0% |
| Legal Research Bench | 48.1% | 38.0% |
| MedCode | 44.7% | 43.3% |
| MedScribe | 86.5% | 86.9% |
| MortgageTax | 64.2% | 61.8% |
| SAGE | 28.9% | 35.0% |
| Tax Agent Bench | 33.1% | 24.9% |
| TaxEval v2 | 71.1% | 71.7% |
| Novel problems | ||
| ARC-AGI-2 | 67.1% | 52.6% |
| ARC-AGI-1 | 87.5% | 87.2% |
| ARC-AGI-3 | 2.1% | 0.3% |
Tests at least two of these models have taken. Bold rows are the core tests.
