Grok 4.6 vs Muse Spark
Grok 4.6 
Muse Spark Verdict
- Muse Spark leads on quality: 65.4 vs 63.8.
- Grok 4.6 (medium) is stronger in composite, knowledge, long context.
- Muse Spark is stronger in coding, reasoning, agents & tools, human preference, instruction following, maths, multimodal.
| Metric | Grok 4.6 (medium) | Muse Spark |
|---|---|---|
| BenchLeader Index | 63.8 | 65.4 |
| Coding score | 63.9 | 66.0 |
| Composite score | 83.2 | 68.4 |
| Knowledge score | 77.3 | 64.5 |
| Long context score | 67.1 | 65.5 |
| Reasoning score | 62.1 | 70.0 |
| Agents & tools score | – | 66.3 |
| Human preference score | – | 69.0 |
| Instruction following score | – | 79.6 |
| Maths score | – | 56.3 |
| Multimodal score | – | 65.9 |
| Blended price $/M | $3.00 | – |
| Output speed | 55 tok/s | – |
| Time to first answer | 36.7 s | – |
| Context window | 500k | 262k |
| GPQA Diamond | – | 89.8% |
| OTIS Mock AIME | – | 88.9% |
| Humanity's Last Exam | – | 40.6% |
| SciCode | 54.6% | 51.5% |
| ProofBench | – | 17.0% |
| Epoch Capabilities Index | – | 152.1 |
| LMArena Text | – | 1488 |
| LMArena Hard Prompts | – | 1505 |
| LMArena Coding | – | 1526 |
| LMArena Vision | – | 1306 |
| AA Intelligence Index | 43.0 | 31.3 |
| IFBench | – | 75.9% |
| AA-LCR | 81.0% | 78.0% |
| MMMU-Pro | – | 80.5% |
| AA-Omniscience | 28 | 7.2 |
| Terminal-Bench Hard | – | 45.5% |
| GPQA Diamond (AA) | 93.5% | 88.4% |
| Humanity's Last Exam (AA) | 42.1% | 40.7% |
| SciCode (AA) | 55.9% | – |
| τ²-Bench Telecom (AA) | – | 91.5% |
| AIME (Vals) | – | 96.9% |
| MMLU-Pro | – | 87.3% |
| LegalBench | – | 84.2% |
| CorpFin | – | 65.1% |
| TaxEval | – | 77.7% |
| SWE-bench (Vals) | – | 74.4% |
| GPQA Diamond (Vals) | – | 89.7% |
| SWE-Bench Pro | – | 55.0% |
| MCP Atlas | – | 82.2% |
| MultiChallenge | – | 75.5% |
| PRBench Finance | – | 52.4% |
| PRBench Legal | – | 52.3% |
| MultiNRC | – | 59.0% |
| TutorBench | – | 68.5% |
| 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.