| Benchmark | Gemini 3.5 Flash | Muse Spark 1.1 |
|---|---|---|
| AA Agentic Index | 70.4 | 39.7 |
| AA Intelligence | 52 | 53.2 |
| AA-LCR | 81 | 81.3 |
| AA-Omniscience | 21.2 | 28.1 |
| arena_elo | 1477 | 1491 |
| arena_text_factuality | 1466 | 1466 |
| Artificial Analysis Coding Index | 70.1 | 71.3 |
| charxiv_reasoning_with_tools | 84.9 | 88.4 |
| coding_arena_elo | 1490 | 1539 |
| critpt | 13.1 | 15.1 |
| DeepSWE 1.1 | 37 | 53.3 |
| Finance Agent v2 | 57.9 | 57.2 |
| gdpval | 57.8 | 43.7 |
| GPQA Diamond | 92.2 | 89.8 |
| HLE | 42.7 | 62.1 |
| livebench_agentic_coding | 49 | 58.5 |
| livebench_coding | 78.2 | 77.2 |
| livebench_data_analysis | 64.9 | 72.5 |
| livebench_instruction_following | 75.6 | 69.6 |
| livebench_language | 84.6 | 74.3 |
| livebench_math | 88.2 | 87.1 |
| livebench_reasoning | 82 | 87.7 |
| MCP Atlas | 83.6 | 88.1 |
| OmniScience Accuracy | 51.4 | 52 |
| OmniScience Non-Hallucination | 38.2 | 50 |
| OSWorld-Verified | 78.4 | 80.8 |
| scicode | 53.1 | 58.2 |
| SWE-bench Pro | 55.1 | 61.5 |
| TauBench V3 - Banking | 32.2 | 31.8 |
| Terminal-Bench 2.1 | 78.7 | 80 |
| toolathlon | 56.5 | 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.