| Benchmark | Gemini 3 Flash Preview | GLM-5.2 Full Open Source |
|---|---|---|
| AA Agentic Index | 49.7 | 45.7 |
| AA Intelligence | 38.7 | 52.6 |
| AA-LCR | 73 | 76.7 |
| AA-Omniscience | 10.1 | 4.4 |
| aime_2026 | 96.7 | 99.2 |
| apexAgents | 27.7 | 33.7 |
| arc_agi_1 | 84.7 | 77 |
| ARC-AGI-2 | 33.6 | 22.8 |
| arena_elo | 1473 | 1471 |
| arena_text_factuality | 1452 | 1460 |
| Artificial Analysis Coding Index | 42.6 | 68.8 |
| coding_arena_elo | 1387 | 1582 |
| critpt | 8.6 | 20.9 |
| gdpval | 35.2 | 50.3 |
| GPQA Diamond | 90.4 | 91.2 |
| HLE | 43.5 | 54.7 |
| hmmt_feb_2026 | 89.4 | 92.5 |
| hmmt_nov_2025 | 93.3 | 94.4 |
| IFBench | 78 | 73.3 |
| MCP Atlas | 62 | 82.6 |
| OmniScience Accuracy | 53.4 | 24.3 |
| OmniScience Non-Hallucination | 7.6 | 73.7 |
| scicode | 50.6 | 50.5 |
| simplebench | 61.1 | 58.8 |
| simpleqa_verified | 67.4 | 38.1 |
| SWE-bench Pro | 49.6 | 62.1 |
| TauBench V3 - Banking | 20.8 | 34.6 |
| Terminal-Bench 2.1 | 58 | 82.7 |
| Terminal-Bench Hard | 38.6 | 50.8 |
| toolathlon | 49.4 | 48.2 |
| τ²-Bench Telecom (AA run) | 80.4 | 99.1 |
| τ³-Bench | 17.5 | 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.