Zhipu Qingyan open-source GLM-4-0414 series models

ZhiPu Qingyan Open-sources GLM-4-0414 Series Models

ZhiPu Qingyan today open-sources the GLM-4-0414 series of models. This series includes the base model GLM-4-32B-Base-0414 (pre-trained on 15T data + optimized by reinforcement learning), the inference model GLM-Z1-32B-0414 (enhanced for mathematical and complex tasks), the rumination model GLM-Z1-Rumination-32B-0414 (handles open-ended questions by integrating search tools), as well as the 9B high-efficiency small model GLM-Z1-9B-0414. The entire series enhances instruction-following, code, and function call capabilities through techniques such as rejection sampling and reinforcement learning. In some benchmarks, it can rival larger-scale models like GPT-4o.

The evaluation results of GLM-4-32B-0414 show that it scored 69.6 points in the BFCL-v3 (Overall) benchmark test, on par with the score of GPT-4o-1120, surpassing competitor models such as DeepSeek-V3-0324 and Qwen2.5-Max.

Github | Hugging Face

📮Contribute ☘️Channel 🌸Chat

via Tech Circle🎗 Zaihua Channel📮 - Telegram Channel