THUDM/GLM-4-32B-0414 vs THUDM/GLM-4.1V-9B-Thinking

THUDM/GLM-4-32B-0414

Z.AI · → text

GLM-4-32B-0414 is a next-generation open-source model with 32 billion parameters, delivering performance comparable to OpenAI’s GPT series and DeepSeek V3/R1. It supports smooth local deployment. The base model was pre-trained on 15T of high-quality data, including a large amount of reasoning-focused synthetic content, setting the stage for advanced reinforcement learning. In the post-training phase, techniques like human preference alignment, rejection sampling, and reinforcement learning were used to improve the model’s ability to follow instructions, generate code, and handle function calls—core skills needed for agent-style tasks. GLM-4-32B-0414 has shown strong results in engineering code, artifact generation, function calling, search-based QA, and report writing—sometimes matching or even surpassing larger models like GPT-4o and DeepSeek-V3 (671B) on specific benchmarks.

THUDM/GLM-4.1V-9B-Thinking

Z.AI · → text

GLM-4.1V-9B-Thinking is an open-source Vision Language Model (VLM) jointly released by Zhipu AI and the KEG Laboratory at Tsinghua University, designed specifically for handling complex multimodal cognitive tasks. Based on the GLM-4-9B-0414 foundation model, it significantly enhances cross-modal reasoning ability and stability by introducing the “Chain-of-Thought” reasoning mechanism and using reinforcement learning strategies. As a lightweight model with 9 billion parameters, it strikes a balance between deployment efficiency and performance. In 28 authoritative benchmark evaluations, it matched or even outperformed the 72-billion-parameter Qwen-2.5-VL-72B model in 18 tasks. The model excels not only in image-text understanding, mathematical and scientific reasoning, and video understanding, but also supports images up to 4K resolution and inputs of arbitrary aspect ratios.