GLM-5.3 Flash: benchmark results
Zhipu's open-weight GLM-5.3-Flash (August 20, 2026), the first natively multimodal model in the GLM-5 series, with 320B total and 18B active parameters and a new hybrid sparse-linear attention architecture. Provider: Zhipu. Released 2026-08-20. Access: Open.
Aggregate Rating 1802 ± 5, rank #62 of 1635 rated models, from 639 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| IndicBankBench - Confusing Intent | 71.7 | Strict pass^3 (%; 53 cases passed out of the 53 cases of the | 100 |
| IndicBankBench - Context Switching | 84.09 | Strict pass^3 (%; 44 cases passed out of the 44 cases of the | 100 |
| LLM Stats (AutomationBench; v1.0.6) | 48.8 | Score (%) | 100 |
| LLM Stats (MMVU) | 80.5 | Score (%) | 100 |
| LLM Stats (MVBench) | 77.8 | Score (%) | 100 |
| LLM Stats (Toolathlon) | 78.4 | Score (%) | 100 |
| ParseBench - Semantic Formatting | 79.49 | Semantic Formatting Score | 100 |
| PureDocBench | 79.69 | Three-track average (0-100): mean of the clean, digitally de | 100 |
| PureDocBench - Digital Degraded | 81.03 | Overall parsing score (0-100): mean of text accuracy (100 ti | 100 |
| PureDocBench - Digital Degraded - Table TEDS | 88.84 | Table Tree-Edit-Distance-based Similarity (0-100) between th | 100 |
| Nejumi 4 - MT-Bench (Japanese) | 98.75 | Score (%) | 98.9 |
| ReLE - Language and Instruction Following | 77 | Accuracy (%) | 98.7 |
Interactive version: theaggregate.ai/model?slug=glm-5-3-flash · How It Works · Data refreshed daily, snapshot 2026-10-15.