Grok 4.5 vs Grok 4.6
Grok 4.5 Verdict
- Grok 4.5 and Grok 4.6 (medium) are level on quality (63.4 vs 63.8).
- Grok 4.5 is stronger in agents & tools, human preference, multimodal, reasoning.
- Grok 4.6 (medium) is stronger in coding, composite, knowledge, long context.
- They cost about the same ($3.00 per 1M blended).
| Metric | Grok 4.5 | Grok 4.6 (medium) |
|---|---|---|
| BenchLeader Index | 63.4 | 63.8 |
| Agents & tools score | 58.6 | – |
| Coding score | 59.9 | 63.9 |
| Composite score | 66.0 | 83.2 |
| Human preference score | 66.9 | – |
| Knowledge score | 76.0 | 77.3 |
| Long context score | 66.2 | 67.1 |
| Multimodal score | 64.8 | – |
| Reasoning score | 68.9 | 62.1 |
| Blended price $/M | $3.00 | $3.00 |
| Output speed | 56 tok/s | 55 tok/s |
| Time to first answer | 12.7 s | 36.7 s |
| Context window | 500k | 500k |
| SimpleBench | 70.0% | – |
| SciCode | – | 54.6% |
| WeirdML | 46.4% | – |
| APEX-Agents | 34.2% | – |
| FrontierCode | 42.4% | – |
| Epoch Capabilities Index | 153.9 | – |
| LMArena Text | 1471 | – |
| LMArena Hard Prompts | 1495 | – |
| LMArena Coding | 1523 | – |
| LMArena WebDev | 1556 | – |
| LMArena Vision | 1291 | – |
| LMArena Agent | 3.9 | – |
| LiveBench | 75.8% | – |
| LiveBench Reasoning | 87.2% | – |
| LiveBench Coding | 68.6% | – |
| LiveBench Agentic Coding | 56.5% | – |
| LiveBench Mathematics | 90.8% | – |
| LiveBench Data Analysis | 73.0% | – |
| LiveBench Language | 82.8% | – |
| AA Intelligence Index | 39.1 | 43.0 |
| AA-LCR | 79.3% | 81.0% |
| MMMU-Pro | 80.4% | – |
| AA-Omniscience | 25.3 | 28 |
| GPQA Diamond (AA) | 93.1% | 93.5% |
| Humanity's Last Exam (AA) | 42.7% | 42.1% |
| SciCode (AA) | 55.0% | 55.9% |
| Kagi LLM Benchmark | 83.5% | – |
| ARC-AGI-1 | – | 87.5% |
| ARC-AGI-2 | – | 61.3% |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.