| Benchmark | DeepSeek-R1 | Gemini 3.5 Flash |
|---|---|---|
| AA Agentic Index | 3.1 | 70.4 |
| AA Intelligence | 18.6 | 52 |
| AA-LCR | 56 | 81 |
| AA-Omniscience | -31.3 | 21.2 |
| arc_agi_1 | 21.2 | 92.5 |
| ARC-AGI-2 | 1.3 | 72.1 |
| Artificial Analysis Coding Index | 24.6 | 70.1 |
| critpt | 0.6 | 13.1 |
| gdpval | 1.5 | 57.8 |
| GPQA Diamond | 81 | 92.2 |
| HLE | 17.7 | 42.7 |
| IFBench | 39 | 76.3 |
| OmniScience Accuracy | 30.7 | 51.4 |
| OmniScience Non-Hallucination | 10.5 | 38.2 |
| scicode | 35.7 | 53.1 |
| simplebench | 40.8 | 76.7 |
| simpleqa_verified | 27.4 | 68.4 |
| TauBench V3 - Banking | 6.4 | 32.2 |
| Terminal-Bench 2.1 | 19.1 | 78.7 |
| Terminal-Bench Hard | 6.1 | 46.2 |
| τ²-Bench Telecom (AA run) | 11.4 | 95.6 |
| τ³-Bench | 6.4 | 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.