| Benchmark | GLM-5.1 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 66 | 14.4 |
| AA Intelligence | 41 | 7.7 |
| AA-LCR | 68 | 23 |
| AA-Omniscience | 0.8 | -68.4 |
| Artificial Analysis Coding Index | 55.8 | 6.7 |
| critpt | 4.6 | 0 |
| gdpval | 49.5 | 13.8 |
| GPQA Diamond | 86.8 | 64.1 |
| HLE | 52.3 | 4.6 |
| HMMT 2025 | 94 | 53.1 |
| IFBench | 76.3 | 36.6 |
| MMLU-Pro | 86 | 73.6 |
| OmniScience Accuracy | 25.2 | 12.2 |
| OmniScience Non-Hallucination | 70.1 | 8.2 |
| scicode | 43.8 | 17.1 |
| Terminal-Bench Hard | 43.2 | 1.5 |
| τ²-Bench Telecom (AA run) | 97.7 | 15.5 |
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.