| Benchmark | DeepSeek-V3 | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 1.6 | 49.7 |
| AA Intelligence | 15.4 | 38.7 |
| AA-LCR | 41.3 | 73 |
| AA-Omniscience | -37.6 | 10.1 |
| AIME 2025 | 51.3 | 99.7 |
| Artificial Analysis Coding Index | 23 | 42.6 |
| critpt | 0 | 8.6 |
| gdpval | 0 | 35.2 |
| GPQA Diamond | 68.4 | 90.4 |
| harmbench | 49.7 | 70 |
| HLE | 5.2 | 43.5 |
| hmmt_feb_2025 | 29.2 | 97.5 |
| IFBench | 41 | 78 |
| livebench | 72.4 | 72.4 |
| mmmlu | 79.4 | 91.8 |
| OmniScience Accuracy | 25.4 | 53.4 |
| OmniScience Non-Hallucination | 23.3 | 7.6 |
| scicode | 35.8 | 50.6 |
| simplebench | 27.2 | 61.1 |
| simpleqa | 27.7 | 68.7 |
| SWE-bench Verified | 42 | 78 |
| TauBench V3 - Banking | 4.7 | 20.8 |
| Terminal-Bench 2.1 | 16.9 | 58 |
| Terminal-Bench Hard | 15.2 | 38.6 |
| vectara_answer_rate | 97.5 | 99.8 |
| vectara_avg_summary_length | 81.7 | 90.2 |
| vectara_factual_consistency | 93.9 | 86.5 |
| vectara_hallucination_rate ↓ | 6.1 | 13.5 |
| τ²-Bench Telecom (AA run) | 47.1 | 80.4 |
| τ³-Bench | 4.7 | 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.