| Benchmark | DeepSeek-V4-Pro | Muse Spark 1.1 |
|---|---|---|
| AA Agentic Index | 63.3 | 39.7 |
| AA Intelligence | 53 | 53.2 |
| AA-LCR | 70 | 81.3 |
| AA-Omniscience | -10.6 | 28.1 |
| arena_elo | 1457 | 1491 |
| arena_text_factuality | 1450 | 1466 |
| Artificial Analysis Coding Index | 59.4 | 71.3 |
| coding_arena_elo | 1582 | 1539 |
| critpt | 13 | 15.1 |
| cybergym | 52.7 | 59 |
| gdpval | 49 | 43.7 |
| GPQA Diamond | 90.5 | 89.8 |
| HLE | 48.2 | 62.1 |
| livebench_agentic_coding | 42.6 | 58.5 |
| livebench_coding | 70 | 77.2 |
| livebench_data_analysis | 74.5 | 72.5 |
| livebench_instruction_following | 62.4 | 69.6 |
| livebench_language | 78.1 | 74.3 |
| livebench_math | 90.7 | 87.1 |
| livebench_reasoning | 82.7 | 87.7 |
| MCP Atlas | 74.2 | 88.1 |
| OmniScience Accuracy | 43 | 52 |
| OmniScience Non-Hallucination | 12.2 | 50 |
| scicode | 50 | 58.2 |
| SWE-bench Pro | 55.4 | 61.5 |
| TauBench V3 - Banking | 30.1 | 31.8 |
| Terminal-Bench 2.1 | 72.1 | 80 |
| toolathlon | 51.8 | 75.6 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.