| Benchmark | DeepSeek-V4-Flash | Muse Spark |
|---|---|---|
| AA Agentic Index | 61.3 | 28.7 |
| AA Intelligence | 51.8 | 44.3 |
| AA-LCR | 74.3 | 77 |
| AA-Omniscience | -14.3 | 7.2 |
| ARC-AGI-2 | 46 | 42.5 |
| arena_elo | 1439 | 1488 |
| arena_text_factuality | 1434 | 1467 |
| Artificial Analysis Coding Index | 69.1 | 58.6 |
| coding_arena_elo | 1431 | 1540 |
| critpt | 16.6 | 11.3 |
| cybergym | 76.7 | 43.5 |
| gdpval | 52.9 | 32.3 |
| GPQA Diamond | 90.8 | 89.5 |
| HLE | 38.6 | 58.4 |
| IFBench | 79.2 | 75.9 |
| itbenchSre | 31.5 | 16.2 |
| MCP Atlas | 69 | 82.2 |
| OmniScience Accuracy | 40.4 | 49.6 |
| OmniScience Non-Hallucination | 15.6 | 26.8 |
| scicode | 49.9 | 51.5 |
| simpleqa_verified | 45 | 66.3 |
| SWE-bench Pro | 52.6 | 55 |
| SWE-bench Verified | 79 | 77.4 |
| TauBench V3 - Banking | 39.4 | 19.6 |
| Terminal-Bench 2.0 | 56.9 | 59 |
| Terminal-Bench 2.1 | 82.7 | 62.2 |
| Terminal-Bench Hard | 38.6 | 45.5 |
| τ²-Bench Telecom (AA run) | 95.6 | 91.5 |
| τ³-Bench | 22.9 | 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.