ChatGLM2 6B: benchmark results
Zhipu's ChatGLM2-6B, the second generation of the bilingual Chinese-English ChatGLM dialogue model, extending context from 2K to 32K tokens and adding Multi-Query Attention for faster inference. Provider: Zhipu. Released 2023-06-25. Access: Open.
Aggregate Rating 1290 ± 11, rank #1498 of 1656 rated models, from 127 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLMsPark - Dictator Game (Multiple Play) | 1106.22 | Elo rating (initial 1000, K=32) | 100 |
| LLMsPark - Dictator Game (Single Play) | 1116.96 | Elo rating (initial 1000, K=32) | 100 |
| MMLU-by-task - Business Ethics | 61 | Accuracy (%) | 89.1 |
| MMLU-by-task - Machine Learning | 43.75 | Accuracy (%) | 86.9 |
| MMLU-by-task - High School Chemistry | 45.32 | Accuracy (%) | 85.7 |
| LLMsPark - Trust Game (Single Play) | 1285.09 | Elo rating (initial 1000, K=32) | 84.6 |
| LLMsPark - Prisoner's Dilemma (Single Play) | 1210.51 | Elo rating (initial 1000, K=32) | 76.9 |
| MMLU-by-task - College Chemistry | 40 | Accuracy (%) | 73.8 |
| MMLU-by-task - Global Facts | 36 | Accuracy (%) | 70 |
| MMLU-by-task - Astronomy | 53.29 | Accuracy (%) | 69.8 |
| MMLU-by-task - Formal Logic | 34.92 | Accuracy (%) | 69.3 |
| MMLU-by-task - Econometrics | 30.7 | Accuracy (%) | 68.2 |
Interactive version: theaggregate.ai/model?slug=chatglm2-6b · How It Works · Data refreshed daily, snapshot 2026-10-20.