| Benchmark | Gemma 4 31B | GLM-5.1 |
|---|---|---|
| AA Agentic Index | 14.4 | 66 |
| AA Intelligence | 29.7 | 41 |
| AA-LCR | 68.3 | 68 |
| AA-Omniscience | -47.9 | 0.8 |
| aime_2026 | 89.2 | 95.8 |
| arena_elo | 1451 | 1468 |
| arena_text_factuality | 1449 | 1458 |
| Artificial Analysis Coding Index | 43.4 | 55.8 |
| coding_arena_elo | 1364 | 1510 |
| critpt | 1.4 | 4.6 |
| gdpval | 15.5 | 49.5 |
| GPQA Diamond | 85.7 | 86.8 |
| HLE | 26.5 | 52.3 |
| HLE (with tools) | 26.5 | 52.3 |
| hmmt_feb_2026 | 77.2 | 89.4 |
| hmmt_nov_2025 | 87.5 | 94 |
| IFBench | 75.6 | 76.3 |
| imo_answer_bench | 74.5 | 83.8 |
| itbenchSre | 37.3 | 40.3 |
| MCP Atlas | 57.2 | 75.6 |
| MMLU-Pro | 85.2 | 86 |
| nl2repo | 15.5 | 42.7 |
| OmniScience Accuracy | 20 | 25.2 |
| OmniScience Non-Hallucination | 18.1 | 70.1 |
| scicode | 43.4 | 43.8 |
| simpleqa_verified | 9.6 | 38.1 |
| SWE-bench Multilingual | 51.7 | 73.3 |
| SWE-bench Pro | 35.7 | 58.4 |
| TauBench V3 - Banking | 14.8 | 13.6 |
| Terminal-Bench 2.0 | 42.9 | 69 |
| Terminal-Bench 2.1 | 43.4 | 63.5 |
| Terminal-Bench Hard | 36.4 | 43.2 |
| tool_decathlon | 21.2 | 40.7 |
| τ²-Bench Telecom (AA run) | 65.5 | 97.7 |
| τ³-Bench | 67.5 | 70.6 |
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.