| Benchmark | Gemma 4 26B A4B | GLM-5.2 Full Open Source |
|---|---|---|
| AA Agentic Index | 32.1 | 45.7 |
| AA Intelligence | 26.1 | 52.6 |
| AA-LCR | 61.7 | 76.7 |
| AA-Omniscience | -50.8 | 4.4 |
| aime_2026 | 88.3 | 99.2 |
| arena_elo | 1438 | 1471 |
| arena_text_factuality | 1445 | 1460 |
| Artificial Analysis Coding Index | 39.3 | 68.8 |
| coding_arena_elo | 1361 | 1582 |
| critpt | 0 | 20.9 |
| gdpval | 25.7 | 50.3 |
| GDPval-AA v2 | 807 | 1514 |
| GPQA Diamond | 82.3 | 91.2 |
| HLE | 19.3 | 54.7 |
| HLE (with tools) | 17.2 | 54.7 |
| hmmt_feb_2026 | 79 | 92.5 |
| hmmt_nov_2025 | 87.5 | 94.4 |
| IFBench | 72.4 | 73.3 |
| imo_answer_bench | 74.3 | 91 |
| itbenchSre | 23.6 | 42.7 |
| MCP Atlas | 50 | 82.6 |
| nl2repo | 11.6 | 48.9 |
| OmniScience Accuracy | 19.1 | 24.3 |
| OmniScience Non-Hallucination | 13.6 | 73.7 |
| scicode | 40.3 | 50.5 |
| SWE-bench Pro | 13.8 | 62.1 |
| TauBench V3 - Banking | 12 | 34.6 |
| Terminal-Bench 2.1 | 39 | 82.7 |
| Terminal-Bench Hard | 25 | 50.8 |
| tool_decathlon | 12 | 48.2 |
| τ²-Bench Telecom (AA run) | 43.6 | 99.1 |
| τ³-Bench | 59 | 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.