| Benchmark | DeepSeek-V4-Pro | Muse Spark |
|---|---|---|
| AA Agentic Index | 63.3 | 28.7 |
| AA Intelligence | 53 | 44.3 |
| AA-LCR | 70 | 77 |
| AA-Omniscience | -10.6 | 7.2 |
| arena_elo | 1457 | 1488 |
| arena_text_factuality | 1450 | 1467 |
| Artificial Analysis Coding Index | 59.4 | 58.6 |
| coding_arena_elo | 1582 | 1540 |
| critpt | 13 | 11.3 |
| cybergym | 52.7 | 43.5 |
| gdpval | 49 | 32.3 |
| GPQA Diamond | 90.5 | 89.5 |
| HLE | 48.2 | 58.4 |
| IFBench | 76.5 | 75.9 |
| itbenchSre | 38.3 | 16.2 |
| MCP Atlas | 74.2 | 82.2 |
| OmniScience Accuracy | 43 | 49.6 |
| OmniScience Non-Hallucination | 12.2 | 26.8 |
| scicode | 50 | 51.5 |
| simpleqa_verified | 57.9 | 66.3 |
| SWE-bench Pro | 55.4 | 55 |
| SWE-bench Verified | 80.6 | 77.4 |
| TauBench V3 - Banking | 30.1 | 19.6 |
| Terminal-Bench 2.0 | 67.9 | 59 |
| Terminal-Bench 2.1 | 72.1 | 62.2 |
| Terminal-Bench Hard | 46.2 | 45.5 |
| τ²-Bench Telecom (AA run) | 96.2 | 91.5 |
| τ³-Bench | 25.8 | 19.6 |
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.