GLM-4.6 (Thinking): benchmark results
GLM-4.6 evaluated with thinking enabled. Provider: Zhipu. Released 2025-09-30. Access: Open.
Aggregate Rating 1643 ± 17, rank #547 of 2168 rated models, from 45 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BenchTable | 72.9 | Total Score (%) | 90.4 |
| BenchTable - Reasoning | 75.3 | Weighted Score (%) | 89.2 |
| AgentNoiseBench (tau2-bench, User Noise) | 68 | Success on a stratified 25 percent sample of tau2-bench task | 89.1 |
| BenchTable - Utility | 72.4 | Weighted Score (%) | 87.8 |
| IF-RewardBench - Overall Assessment (Multi-Turn) | 0.26 | Kendall tau-b (-1 to 1) between the judge ranking of the res | 87.5 |
| BenchTable - STEM | 75.1 | Weighted Score (%) | 85.2 |
| AgentNoiseBench (VitaBench, Tool Noise) | 40 | Success on a stratified 25 percent sample of VitaBench tasks | 84.8 |
| AgentNoiseBench (VitaBench, User Noise) | 40 | Success on a stratified 25 percent sample of VitaBench tasks | 84.8 |
| AgentNoiseBench (tau2-bench, Tool Noise) | 56 | Success on a stratified 25 percent sample of tau2-bench task | 84.8 |
| LLM2014 Code 2025-11 - C# | 8.38 | Score | 84 |
| IF-RewardBench - Overall Assessment | 0.27 | Kendall tau-b (-1 to 1) between the judge ranking of the res | 81.2 |
| IF-RewardBench - Overall Assessment (Single-Turn) | 0.36 | Kendall tau-b (-1 to 1) between the judge ranking of the res | 81.2 |
Interactive version: theaggregate.ai/model?slug=glm-4-6-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.