GLM-4.6: benchmark results
Zhipu's open GLM-4.6 model with a 200K context, improving agentic coding over GLM-4.5. Provider: Zhipu. Released 2025-09-30. Access: Open.
Aggregate Rating 1678 ± 5, rank #252 of 1656 rated models, from 859 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentNoiseBench (Search, Tool Noise) | 39 | Accuracy on 500 sampled 2WikiMultiHopQA and HotpotQA questio | 100 |
| DDR-Bench - GLOBEM - Item-Averaged Message-Wise | 41.61 | Checklist Coverage Accuracy (%) | 100 |
| DDR-Bench - GLOBEM - Sample-Averaged Message-Wise | 41.56 | Checklist Coverage Accuracy (%) | 100 |
| Diplomacy: Steerability | 32.5 | Mean game-score difference: optimized aggressive prompt minu | 100 |
| DocFormBench (DocFormFlow) - Hallucinated Formatting | 1.24 | Hallucinated formatting rate (%) on DocFormBench: share of a | 100 |
| FinReasoning (Deep Insight) - Factuality and Contextualization | 88.1 | Rubric score (0-100): a judge score fused from DeepSeek-V3 a | 100 |
| LLMEval-Fair - Management | 9.63 | Discipline Score (0-10) | 98.3 |
| OpenCompass Knowledge - Social Science - Chinese (CompassBench 2510) | 94 | Score (%) | 98.1 |
| AGC-Bench - twistlist | 1.95 | Dataset z-score | 97.6 |
| Berkeley Function-Calling Leaderboard | 72.38 | Overall Accuracy (%) | 97.2 |
| Gorilla API Bench (BFCL) | 72.38 | Overall Accuracy (%) | 97.2 |
| SalesLLM - Chinese (GPT-4o User) | 6.74 | SalesLLM benchmark score (1-10): 0.5 x buying-intent score ( | 96.4 |
Interactive version: theaggregate.ai/model?slug=glm-4-6 · How It Works · Data refreshed daily, snapshot 2026-10-20.