| Benchmark | Gemma 4 31B | GLM-5.2 Full Open Source |
|---|---|---|
| AA Agentic Index | 14.4 | 45.7 |
| AA Intelligence | 29.7 | 52.6 |
| AA-LCR | 68.3 | 76.7 |
| AA-Omniscience | -47.9 | 4.4 |
| aime_2026 | 89.2 | 99.2 |
| arena_elo | 1451 | 1471 |
| arena_text_factuality | 1449 | 1460 |
| Artificial Analysis Coding Index | 43.4 | 68.8 |
| coding_arena_elo | 1364 | 1582 |
| critpt | 1.4 | 20.9 |
| gdpval | 15.5 | 50.3 |
| GPQA Diamond | 85.7 | 91.2 |
| HLE | 26.5 | 54.7 |
| HLE (with tools) | 26.5 | 54.7 |
| hmmt_feb_2026 | 77.2 | 92.5 |
| hmmt_nov_2025 | 87.5 | 94.4 |
| IFBench | 75.6 | 73.3 |
| imo_answer_bench | 74.5 | 91 |
| itbenchSre | 37.3 | 42.7 |
| MCP Atlas | 57.2 | 82.6 |
| nl2repo | 15.5 | 48.9 |
| OmniScience Accuracy | 20 | 24.3 |
| OmniScience Non-Hallucination | 18.1 | 73.7 |
| scicode | 43.4 | 50.5 |
| simpleqa_verified | 9.6 | 38.1 |
| SWE-bench Pro | 35.7 | 62.1 |
| TauBench V3 - Banking | 14.8 | 34.6 |
| Terminal-Bench 2.1 | 43.4 | 82.7 |
| Terminal-Bench Hard | 36.4 | 50.8 |
| tool_decathlon | 21.2 | 48.2 |
| τ²-Bench Telecom (AA run) | 65.5 | 99.1 |
| τ³-Bench | 67.5 | 26.8 |
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.