| Benchmark | Claude Opus 4.8 | Gemma 4 26B A4B |
|---|---|---|
| AA Agentic Index | 49.4 | 32.1 |
| AA Intelligence | 57.3 | 26.1 |
| AA-LCR | 73 | 61.7 |
| AA-Omniscience | 28.8 | -50.8 |
| aime_2026 | 100 | 88.3 |
| arena_elo | 1482 | 1438 |
| arena_text_factuality | 1464 | 1445 |
| arena_vision | 1280 | 1238 |
| Artificial Analysis Coding Index | 74.3 | 39.3 |
| browsecomp | 88.5 | 26.3 |
| charxiv_rq | 80.5 | 69 |
| coding_arena_elo | 1539 | 1361 |
| critpt | 20.9 | 0 |
| gdpval | 54.2 | 25.7 |
| GDPval-AA v2 | 1593 | 807 |
| GPQA Diamond | 93.6 | 82.3 |
| HLE | 57.9 | 19.3 |
| HLE (with tools) | 57.9 | 17.2 |
| hmmt_feb_2026 | 96.7 | 79 |
| hmmt_nov_2025 | 96.5 | 87.5 |
| IFBench | 62.2 | 72.4 |
| imo_answer_bench | 83.5 | 74.3 |
| mathvision | 86.7 | 82.4 |
| MCP Atlas | 83.6 | 50 |
| MMMU-Pro | 78.9 | 73.8 |
| nl2repo | 69.7 | 11.6 |
| OmniScience Accuracy | 48.8 | 19.1 |
| OmniScience Non-Hallucination | 60.7 | 13.6 |
| scicode | 53.5 | 40.3 |
| SWE-bench Multilingual | 84.4 | 43.4 |
| SWE-bench Pro | 69.2 | 13.8 |
| SWE-bench Verified | 88.6 | 57.4 |
| TauBench V3 - Banking | 34.2 | 12 |
| Terminal-Bench 2.0 | 74.6 | 34.2 |
| Terminal-Bench 2.1 | 85 | 39 |
| Terminal-Bench Hard | 58.3 | 25 |
| tool_decathlon | 59.9 | 12 |
| WideSearch | 72.9 | 38.3 |
| τ²-Bench Telecom (AA run) | 94.4 | 43.6 |
| τ³-Bench | 27.6 | 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.