| Benchmark | Gemini 3.5 Flash | GLM-5.2 Full Open Source |
|---|---|---|
| AA Agentic Index | 70.4 | 45.7 |
| AA Intelligence | 52 | 52.6 |
| AA-LCR | 81 | 76.7 |
| AA-Omniscience | 21.2 | 4.4 |
| aime_2026 | 95 | 99.2 |
| apexAgents | 47.1 | 33.7 |
| arc_agi_1 | 92.5 | 77 |
| ARC-AGI-2 | 72.1 | 22.8 |
| arena_elo | 1477 | 1471 |
| arena_text_factuality | 1466 | 1460 |
| Artificial Analysis Coding Index | 70.1 | 68.8 |
| coding_arena_elo | 1490 | 1582 |
| critpt | 13.1 | 20.9 |
| DeepSWE | 37 | 46.2 |
| DeepSWE 1.1 | 37 | 44 |
| gdpval | 57.8 | 50.3 |
| GDPval-AA v2 | 1357 | 1514 |
| GPQA Diamond | 92.2 | 91.2 |
| HiL-Bench | 27.7 | 43.7 |
| HLE | 42.7 | 54.7 |
| hmmt_feb_2026 | 95.5 | 92.5 |
| IFBench | 76.3 | 73.3 |
| itbenchSre | 40.3 | 42.7 |
| livebench_agentic_coding | 49 | 51.8 |
| livebench_coding | 78.2 | 79.7 |
| livebench_data_analysis | 64.9 | 73.7 |
| livebench_instruction_following | 75.6 | 62.3 |
| livebench_language | 84.6 | 76.2 |
| livebench_math | 88.2 | 89.8 |
| livebench_reasoning | 82 | 78.6 |
| MCP Atlas | 83.6 | 82.6 |
| OmniScience Accuracy | 51.4 | 24.3 |
| OmniScience Non-Hallucination | 38.2 | 73.7 |
| scicode | 53.1 | 50.5 |
| simplebench | 76.7 | 58.8 |
| simpleqa_verified | 68.4 | 38.1 |
| SWE-bench Pro | 55.1 | 62.1 |
| TauBench V3 - Banking | 32.2 | 34.6 |
| Terminal-Bench 2.1 | 78.7 | 82.7 |
| Terminal-Bench Hard | 46.2 | 50.8 |
| toolathlon | 56.5 | 48.2 |
| τ²-Bench Telecom (AA run) | 95.6 | 99.1 |
| τ³-Bench | 25.4 | 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.