GLM-5.3-FlashX-Legal: benchmark results
Z.ai's legal-specialized GLM-5.3-FlashX-Legal, which Z.ai's documentation also calls GLM-Legal. Z.ai announced GLM-Legal on October 12, 2026 as a beta built on GLM-5.3-Flash with legal mid-training and reinforcement learning, and released it in AutoClaw under the full name in late October. Z.ai reports a 19.2% all-pass rate on Harvey LAB, 12.5 points above its base model, in an evaluation run by Vals. No downloadable checkpoint has been verified. Provider: Zhipu. Released 2026-10-12. Access: API.
Aggregate Rating 1843 ± 26, rank #43 of 10916 rated models, from 11 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLM Stats (Legal Agent Benchmark) | 19.2 | Score (%) | 100 |
| Vals AI Harvey Legal Agent Bench | 19.17 | Accuracy (%) | 92.5 |
| Vals AI LegalBench | 86.24 | Accuracy (%) | 91.4 |
| Tinybird AI SQL Benchmark - Exactness | 55.29 | Result exactness vs human reference queries (0-100) | 89 |
| DuelLab GameBench 2 - Reasoning Variants | 69.75 | GameBench2 normalized score (0–100) | 82.9 |
| LLM Stats Score | 41.48 | LLM Stats Score (conservative rating) | 82.8 |
| GLM-Legal Release - PLawBench | 77.3 | Reported rubric score | 75 |
| Tinybird AI SQL Benchmark - Success Rate | 100 | Questions answered with a valid query within 3 retries (%) | 72 |
| Vals AI Legal Research Bench | 41.83 | Accuracy (%) | 70.6 |
| Tinybird AI SQL Benchmark - First-Attempt Success Rate | 94 | Questions answered with a valid query on the first attempt ( | 52.7 |
| LM Market Cap LMC Score | 69.7 | LMC Score (0-100) | 42.6 |
Interactive version: theaggregate.ai/model?slug=glm-5-3-flashx-legal · How It Works · Data refreshed daily, snapshot 2026-11-08.