| Benchmark | DeepSeek-V3.2 | Gemini 3.5 Flash |
|---|---|---|
| AA Agentic Index | 39.8 | 70.4 |
| AA Intelligence | 32.8 | 52 |
| AA-LCR | 70.7 | 81 |
| AA-Omniscience | -22.5 | 21.2 |
| aime_2026 | 95.1 | 95 |
| apexAgents | 14.5 | 47.1 |
| arc_agi_1 | 57 | 92.5 |
| ARC-AGI-2 | 4 | 72.1 |
| arena_text_factuality | 1433 | 1466 |
| Artificial Analysis Coding Index | 44.2 | 70.1 |
| coding_arena_elo | 1368 | 1490 |
| critpt | 2.9 | 13.1 |
| frontiermath_tier_4 | 2.1 | 14.6 |
| gdpval | 18.8 | 57.8 |
| GPQA Diamond | 84 | 92.2 |
| HLE | 40.8 | 42.7 |
| hmmt_feb_2026 | 84.1 | 95.5 |
| IFBench | 61 | 76.3 |
| MCP Atlas | 62.2 | 83.6 |
| OmniScience Accuracy | 33 | 51.4 |
| OmniScience Non-Hallucination | 17.3 | 38.2 |
| scicode | 39 | 53.1 |
| simpleqa_verified | 27.5 | 68.4 |
| SWE-bench Pro | 15.6 | 55.1 |
| TauBench V3 - Banking | 18.8 | 32.2 |
| Terminal-Bench 2.0 | 46.4 | 76.2 |
| Terminal-Bench 2.1 | 46.8 | 78.7 |
| Terminal-Bench Hard | 35.6 | 46.2 |
| toolathlon | 35.2 | 56.5 |
| τ²-Bench Telecom (AA run) | 90.6 | 95.6 |
| τ³-Bench | 69.2 | 25.4 |
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.