| Benchmark | DeepSeek-V4-Pro | Gemini 3.5 Flash |
|---|---|---|
| AA Agentic Index | 63.3 | 70.4 |
| AA Intelligence | 53 | 52 |
| AA-LCR | 70 | 81 |
| AA-Omniscience | -10.6 | 21.2 |
| aime_2026 | 96.7 | 95 |
| apexAgents | 24.3 | 47.1 |
| arena_elo | 1457 | 1477 |
| arena_text_factuality | 1450 | 1466 |
| Artificial Analysis Coding Index | 59.4 | 70.1 |
| coding_arena_elo | 1582 | 1490 |
| critpt | 13 | 13.1 |
| DeepSWE | 12.8 | 37 |
| gdpval | 49 | 57.8 |
| GDPval-AA (Elo) | 1554 | 1656 |
| GDPval-AA v2 | 1307 | 1357 |
| GPQA Diamond | 90.5 | 92.2 |
| HLE | 48.2 | 42.7 |
| hmmt_feb_2026 | 95.2 | 95.5 |
| IFBench | 76.5 | 76.3 |
| itbenchSre | 38.3 | 40.3 |
| livebench | 73.6 | 75 |
| livebench_agentic_coding | 42.6 | 49 |
| livebench_coding | 70 | 78.2 |
| livebench_data_analysis | 74.5 | 64.9 |
| livebench_instruction_following | 62.4 | 75.6 |
| livebench_language | 78.1 | 84.6 |
| livebench_math | 90.7 | 88.2 |
| livebench_reasoning | 82.7 | 82 |
| MCP Atlas | 74.2 | 83.6 |
| OmniScience Accuracy | 43 | 51.4 |
| OmniScience Non-Hallucination | 12.2 | 38.2 |
| scicode | 50 | 53.1 |
| simplebench | 50.9 | 76.7 |
| simpleqa_verified | 57.9 | 68.4 |
| SWE-bench Pro | 55.4 | 55.1 |
| TauBench V3 - Banking | 30.1 | 32.2 |
| Terminal-Bench 2.0 | 67.9 | 76.2 |
| Terminal-Bench 2.1 | 72.1 | 78.7 |
| Terminal-Bench Hard | 46.2 | 46.2 |
| toolathlon | 51.8 | 56.5 |
| τ²-Bench Telecom (AA run) | 96.2 | 95.6 |
| τ³-Bench | 25.8 | 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.