| Benchmark | DeepSeek-V4-Pro | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 63.3 | 49.7 |
| AA Intelligence | 53 | 38.7 |
| AA-LCR | 70 | 73 |
| AA-Omniscience | -10.6 | 10.1 |
| aime_2026 | 96.7 | 96.7 |
| apexAgents | 24.3 | 27.7 |
| arena_elo | 1457 | 1473 |
| arena_text_factuality | 1450 | 1452 |
| Artificial Analysis Coding Index | 59.4 | 42.6 |
| coding_arena_elo | 1582 | 1387 |
| critpt | 13 | 8.6 |
| gdpval | 49 | 35.2 |
| GDPval-AA (Elo) | 1554 | 1204 |
| GPQA Diamond | 90.5 | 90.4 |
| HLE | 48.2 | 43.5 |
| hmmt_feb_2026 | 95.2 | 89.4 |
| hmmt_nov_2025 | 94.4 | 93.3 |
| IFBench | 76.5 | 78 |
| livebench | 73.6 | 72.4 |
| MCP Atlas | 74.2 | 62 |
| OmniScience Accuracy | 43 | 53.4 |
| OmniScience Non-Hallucination | 12.2 | 7.6 |
| scicode | 50 | 50.6 |
| simplebench | 50.9 | 61.1 |
| simpleqa | 57.9 | 68.7 |
| simpleqa_verified | 57.9 | 67.4 |
| SWE-bench Multilingual | 76.2 | 72.7 |
| SWE-bench Pro | 55.4 | 49.6 |
| SWE-bench Verified | 80.6 | 78 |
| TauBench V3 - Banking | 30.1 | 20.8 |
| Terminal-Bench 2.0 | 67.9 | 47.6 |
| Terminal-Bench 2.1 | 72.1 | 58 |
| Terminal-Bench Hard | 46.2 | 38.6 |
| toolathlon | 51.8 | 49.4 |
| vectara_answer_rate | 97.2 | 99.8 |
| vectara_avg_summary_length | 153.8 | 90.2 |
| vectara_factual_consistency | 91.4 | 86.5 |
| vectara_hallucination_rate ↓ | 8.6 | 13.5 |
| τ²-Bench Telecom (AA run) | 96.2 | 80.4 |
| τ³-Bench | 25.8 | 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.