| Benchmark | DeepSeek-V3.2 | Gemini 3.6 Flash |
|---|---|---|
| AA Agentic Index | 39.8 | 40.5 |
| AA Intelligence | 32.8 | 52 |
| AA-LCR | 70.7 | 79 |
| AA-Omniscience | -22.5 | 22.1 |
| aime_2026 | 95.1 | 96.7 |
| arc_agi_1 | 57 | 34.5 |
| ARC-AGI-2 | 4 | 2.6 |
| arena_text_factuality | 1433 | 1470 |
| Artificial Analysis Coding Index | 44.2 | 69.2 |
| coding_arena_elo | 1368 | 1538 |
| critpt | 2.9 | 10.6 |
| gdpval | 18.8 | 46.1 |
| GPQA Diamond | 84 | 92.8 |
| HLE | 40.8 | 40.8 |
| hmmt_feb_2026 | 84.1 | 89.4 |
| OmniScience Accuracy | 33 | 50 |
| OmniScience Non-Hallucination | 17.3 | 44.4 |
| scicode | 39 | 52.7 |
| simpleqa_verified | 27.5 | 68.7 |
| SWE-bench Pro | 15.6 | 58.7 |
| TauBench V3 - Banking | 18.8 | 29.9 |
| Terminal-Bench 2.1 | 46.8 | 78 |
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.