GLM-5.3 Flash (Medium): benchmark results
Provider: Zhipu. Released 2026-08-20. Access: Open.
Aggregate Rating 1584 ± 23, rank #787 of 2201 rated models, from 25 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PetriBench - Deadlock (Hard) | 52 | Exact-Match Accuracy (%) | 52.9 |
| PetriBench - L0-Liveness (Hard) | 50.5 | Exact-Match Accuracy (%) | 50 |
| Continual-ARC First Sight - ARC-AGI-1 All | 26.8 | First-sight exact-match solve rate (%) | 40 |
| Continual-ARC First Sight - ARC-AGI-1 Held Out | 24 | First-sight exact-match solve rate (%) | 40 |
| Continual-ARC First Sight - Private All | 26.5 | First-sight exact-match solve rate (%) | 40 |
| Continual-ARC First Sight - Private Held Out | 30 | First-sight exact-match solve rate (%) | 40 |
| WikiTrace GLM Agent Evaluation - PlanCraft Unseen | 55.3 | Three-trial mean native task success (%) | 40 |
| PetriBench - ReachableMarkings (Hard) | 0.2 | Exact-Match Accuracy (%) | 30 |
| PetriBench - MinimumTokenSteps (Hard) | 0.2 | Exact-Match Accuracy (%) | 27.1 |
| PetriBench - Hard | 24.9 | Exact-Match Accuracy (%) | 25.7 |
| PetriBench - L4-Liveness (Hard) | 48 | Exact-Match Accuracy (%) | 21.4 |
| Continual-ARC First Sight - ARC-AGI-1 Stream | 29.5 | First-sight exact-match solve rate (%) | 20 |
Interactive version: theaggregate.ai/model?slug=glm-5-3-flash-medium · How It Works · Data refreshed daily, snapshot 2026-11-12.