| Benchmark | DeepSeek-V4-Pro | Gemma 4 26B A4B |
|---|---|---|
| AA Agentic Index | 63.3 | 32.1 |
| AA Intelligence | 53 | 26.1 |
| AA-LCR | 70 | 61.7 |
| AA-Omniscience | -10.6 | -50.8 |
| aime_2026 | 96.7 | 88.3 |
| arena_elo | 1457 | 1438 |
| arena_text_factuality | 1450 | 1445 |
| Artificial Analysis Coding Index | 59.4 | 39.3 |
| browsecomp | 83.4 | 26.3 |
| coding_arena_elo | 1582 | 1361 |
| critpt | 13 | 0 |
| gdpval | 49 | 25.7 |
| GDPval-AA v2 | 1307 | 807 |
| GPQA Diamond | 90.5 | 82.3 |
| HLE | 48.2 | 19.3 |
| HLE (with tools) | 48.2 | 17.2 |
| hmmt_feb_2026 | 95.2 | 79 |
| hmmt_nov_2025 | 94.4 | 87.5 |
| IFBench | 76.5 | 72.4 |
| imo_answer_bench | 89.8 | 74.3 |
| itbenchSre | 38.3 | 23.6 |
| LiveCodeBench | 93.5 | 79.8 |
| MCP Atlas | 74.2 | 50 |
| MMLU-Pro | 87.5 | 85.2 |
| nl2repo | 38.5 | 11.6 |
| OmniScience Accuracy | 43 | 19.1 |
| OmniScience Non-Hallucination | 12.2 | 13.6 |
| scicode | 50 | 40.3 |
| SWE-bench Multilingual | 76.2 | 43.4 |
| SWE-bench Pro | 55.4 | 13.8 |
| SWE-bench Verified | 80.6 | 57.4 |
| TauBench V3 - Banking | 30.1 | 12 |
| Terminal-Bench 2.0 | 67.9 | 34.2 |
| Terminal-Bench 2.1 | 72.1 | 39 |
| Terminal-Bench Hard | 46.2 | 25 |
| tool_decathlon | 52.8 | 12 |
| τ²-Bench Telecom (AA run) | 96.2 | 43.6 |
| τ³-Bench | 25.8 | 59 |
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.