| Benchmark | DeepSeek-R1 | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 3.1 | 49.7 |
| AA Intelligence | 18.6 | 38.7 |
| AA-LCR | 56 | 73 |
| AA-Omniscience | -31.3 | 10.1 |
| AIME 2025 | 87.5 | 99.7 |
| arc_agi_1 | 21.2 | 84.7 |
| ARC-AGI-2 | 1.3 | 33.6 |
| Artificial Analysis Coding Index | 24.6 | 42.6 |
| critpt | 0.6 | 8.6 |
| gdpval | 1.5 | 35.2 |
| GPQA Diamond | 81 | 90.4 |
| harmbench | 54.6 | 70 |
| HLE | 17.7 | 43.5 |
| hmmt_feb_2025 | 76.7 | 97.5 |
| IFBench | 39 | 78 |
| OmniScience Accuracy | 30.7 | 53.4 |
| OmniScience Non-Hallucination | 10.5 | 7.6 |
| scicode | 35.7 | 50.6 |
| simplebench | 40.8 | 61.1 |
| simpleqa | 92.3 | 68.7 |
| simpleqa_verified | 27.4 | 67.4 |
| SWE-bench Multilingual | 30.5 | 72.7 |
| SWE-bench Verified | 57.6 | 78 |
| TauBench V3 - Banking | 6.4 | 20.8 |
| Terminal-Bench 2.1 | 19.1 | 58 |
| Terminal-Bench Hard | 6.1 | 38.6 |
| vectara_answer_rate | 97 | 99.8 |
| vectara_avg_summary_length | 93.5 | 90.2 |
| vectara_factual_consistency | 88.7 | 86.5 |
| vectara_hallucination_rate ↓ | 11.3 | 13.5 |
| τ²-Bench Telecom (AA run) | 11.4 | 80.4 |
| τ³-Bench | 6.4 | 17.5 |
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.