| Benchmark | Claude 4 Opus | Gemma 4 26B A4B |
|---|---|---|
| AA Intelligence | 31.7 | 26.1 |
| AA-LCR | 40 | 61.7 |
| AIME 2025 | 75.5 | 90 |
| arena_vision | 1207 | 1238 |
| Artificial Analysis Coding Index | 34 | 39.3 |
| GPQA Diamond | 79.6 | 82.3 |
| HLE | 12.3 | 19.3 |
| hmmt_feb_2025 | 60 | 91.7 |
| IFBench | 53.7 | 72.4 |
| ifeval | 87.4 | 96.4 |
| LiveCodeBench | 70.4 | 79.8 |
| LiveCodeBench v6 | 47.4 | 77.1 |
| mmlu | 92.9 | 77.8 |
| mmlu_redux | 94.2 | 92.7 |
| MMLU-Pro | 86.6 | 85.2 |
| mmmlu | 88.8 | 86.3 |
| MMMU | 73.7 | 78.4 |
| scicode | 40.9 | 40.3 |
| supergpqa | 56.5 | 61.4 |
| SWE-bench Verified | 79.4 | 57.4 |
| Terminal-Bench Hard | 31.1 | 25 |
| τ²-Bench | 57 | 68.2 |
| τ²-Bench Telecom (AA run) | 73.4 | 43.6 |
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.