| Benchmark | DeepSeek-V3.2 | Muse Spark 1.1 |
|---|---|---|
| AA Agentic Index | 39.8 | 39.7 |
| AA Intelligence | 32.8 | 53.2 |
| AA-LCR | 70.7 | 81.3 |
| AA-Omniscience | -22.5 | 28.1 |
| arena_text_factuality | 1433 | 1466 |
| Artificial Analysis Coding Index | 44.2 | 71.3 |
| coding_arena_elo | 1368 | 1539 |
| critpt | 2.9 | 15.1 |
| cybergym | 17.3 | 59 |
| deepsearchqa_f1 | 60.9 | 84.9 |
| gdpval | 18.8 | 43.7 |
| GPQA Diamond | 84 | 89.8 |
| HLE | 40.8 | 62.1 |
| MCP Atlas | 62.2 | 88.1 |
| OmniScience Accuracy | 33 | 52 |
| OmniScience Non-Hallucination | 17.3 | 50 |
| scicode | 39 | 58.2 |
| SWE-bench Pro | 15.6 | 61.5 |
| TauBench V3 - Banking | 18.8 | 31.8 |
| Terminal-Bench 2.1 | 46.8 | 80 |
| toolathlon | 35.2 | 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.