GLM-4.7 is a large language model engineered specifically for task-oriented development and complex agent workflows. With 355 billion total parameters and 32 billion active parameters, it brings substantial compute to bear on coding, reasoning, and tool-use challenges. The model builds on its predecessor GLM-4.6 and introduces advanced interleaved thinking and preserved thinking capabilities, allowing it to pause and deliberate before taking action in complex multi-step tasks. This design philosophy makes it particularly strong for development environments that involve lengthy task cycles, frequent tool calls, and higher demands for consistency and stability.

The model demonstrates meaningful improvements across software engineering benchmarks, achieving 73.8% on SWE-bench Verified with a +5.8% gain over its predecessor, and 66.7% on the multilingual variant with a +12.9% improvement. Mathematical and reasoning capabilities also see substantial gains, reaching 42.8% on the HLE benchmark with a +12.4% improvement. GLM-4.7 ranks as the top open-source model on LMArena Code Arena and ranks sixth overall in web development benchmarks among all models. It produces cleaner, more modern front-end output with improved layout and sizing, and shows significant improvements when integrated with popular agent frameworks such as Claude Code, Kilo Code, Cline, and Roo Code. The model is available with open weights, making it accessible for developers who want to deploy, fine-tune, or study its behavior in production environments.