| Benchmark | DeepSeek-V3.2 | Gemma 4 E2B |
|---|---|---|
| AA Agentic Index | 39.8 | 7.4 |
| AA Intelligence | 32.8 | 10 |
| AA-LCR | 70.7 | 17 |
| AA-Omniscience | -22.5 | -23.6 |
| aime_2026 | 95.1 | 37.5 |
| AIME25 no tools | 89.3 | 26.3 |
| Artificial Analysis Coding Index | 44.2 | 8.3 |
| critpt | 2.9 | 0 |
| gdpval | 18.8 | 0 |
| GPQA Diamond | 84 | 43.4 |
| HLE | 40.8 | 4.8 |
| IFBench | 61 | 38 |
| LiveCodeBench v6 | 83.3 | 44 |
| LiveCodeBenchV6 no tools | 74.1 | 54.9 |
| MMLU-Pro | 86 | 60 |
| OmniScience Accuracy | 33 | 8.2 |
| OmniScience Non-Hallucination | 17.3 | 67.6 |
| scicode | 39 | 20.9 |
| TauBench V3 - Banking | 18.8 | 4.5 |
| Terminal-Bench 2.1 | 46.8 | 0.4 |
| Terminal-Bench Hard | 35.6 | 3 |
| τ²-Bench | 91 | 24.5 |
| τ²-Bench Telecom (AA run) | 90.6 | 22.2 |
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.