GLM-4.5V: benchmark results
GLM-4.5V is a publisher-released checkpoint supporting a published 65,536-token context; endpoint limits may differ. Provider: Zhipu. Released 2025-08-11. Access: Open.
Aggregate Rating 1633 ± 5, rank #346 of 1656 rated models, from 558 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AndroidIntent - Complete Instructions | 54.5 | Cumulative success rate (%; offline step-wise execution of 7 | 100 |
| AutoGUI-v2 - Region Grounding (Description) | 93.7 | Region grounding by layout description on the same regions, | 100 |
| AutoGUI-v2 - Region Grounding (Functionality) | 84.6 | Region grounding by functionality on 442 tasks, grounding ce | 100 |
| CFMME - Seal Detection | 37.45 | Seal detection mAP (%): boxes and categories (company or per | 100 |
| CareQA-Vision - Open-Ended - Medicine | 68.06 | Judge score (%; 108 open-ended image-based questions from MI | 100 |
| CareQA-Vision - Open-Ended - Nursing | 51.79 | Judge score (%; 28 open-ended image-based questions from EIR | 100 |
| ENC-Bench - Safety Passage Assessment | 65.67 | Accuracy (%) on ENC-Bench's 702 safety passage assessment qu | 100 |
| M-JudgeBench - Incidental Mistake Detection | 99.21 | pairwise accuracy (%) on the 252 pairs where the rejected ch | 100 |
| M-JudgeBench - Logical Fallacy Detection | 99.58 | pairwise accuracy (%) on the 238 pairs where the rejected ch | 100 |
| SPD-Faith Bench - Category-Level F1 | 69.6 | Micro F1 (%) of the COCO object categories the model reports | 100 |
| SPD-Faith Bench - Color Change F1 | 76.7 | F1 (%) for reporting the colour-change edits, on the 1,000 m | 100 |
| SPD-Faith Bench - Difference Reasoning Faithfulness | 58.3 | Share (%) of the differences the model describes, after bein | 100 |
Interactive version: theaggregate.ai/model?slug=glm-4-5v · How It Works · Data refreshed daily, snapshot 2026-10-20.