GLM-5 is Zhipu AI's new-generation flagship foundation model, specifically designed for Coding and Agent scenarios. It achieves State-Of-The-Art (SOTA) performance in open-source complex system engineering and long-horizon tasks, with a real-world coding experience approaching Claude Opus level. Based on a 744B scale foundation model, combined with asynchronous reinforcement learning and sparse attention mechanisms, GLM-5 marks a paradigm shift from "writing code" to "building systems".
- Parameter Scale and Data Volume: The base model's parameter scale has expanded to 744B (with 40B activated parameters), and pre-training data has increased from 23T to 28.5T, significantly enhancing the model's breadth and depth of knowledge.
- Ultra-Long Context and Output: Supports a context window of up to 200K tokens and a maximum output length of 128K tokens, enabling excellent performance in handling long texts, complex code, and multi-step tasks.
- Exceptional Coding and Agent Capabilities: Systematically strengthened programming and agent capabilities, excelling in code generation, low hallucination rates, and long-text processing, capable of more precise search iterations and more efficient token utilization.
- Multiple Thinking Modes: Offers various thinking modes to support more flexible and in-depth problem-solving.
- Complex System Engineering: Suitable for scenarios requiring the construction and management of complex software systems, assisting in system design, code generation, and optimization.
- Long-Horizon Agent Tasks: Excels in agent tasks requiring multi-step planning, execution, and feedback, such as automated workflows and intelligent assistants.
- High-Precision Code Generation and Debugging: Provides human-level coding assistance for developers, improving development efficiency and code quality.
- Large-Scale Document Analysis and Summarization: With its ultra-long context window, it can process and understand massive amounts of documents for deep information extraction and summarization.
| Capability | Detailed Description |
|---|
| Reasoning Ability | Extremely Strong. Excels in complex logical reasoning, multi-step task planning, and execution. |
| Creative Ability | Extremely Strong. Particularly adept at code generation, system design, and innovative problem-solving. |
| Multimodal Ability | Zhipu AI platform offers visual model support; GLM-5 primarily focuses on text and code but can be integrated with multimodal tools. |
| Response Speed | 30-50 tokens/s. Provides relatively fast response speed while ensuring quality. |
| Context Window | 200K |
| Max Output | 128K |
| Model | Input (Credits/Token) | Output (Credits/Token) |
|---|
| GLM-5 | 0.30 | 2.55 |
