Model Releases
GLM-4.7 Release: Zhipu AI Open-Source Coding Model
Zhipu AI releases GLM-4.7, an open-weights coding model topping leaderboards with a cost-effective Flash variant.
December 1, 2025
Model ReleaseGLM-4.7
Introduction
Zhipu AI has officially launched the GLM-4.7 model on December 1, 2025, marking a significant milestone for the open-source community. This release democratizes access to cutting-edge reasoning capabilities, directly challenging established market leaders in both coding and general reasoning tasks. As an open-weights model, GLM-4.7 allows developers to deploy locally or via API without the enterprise licensing costs associated with Western giants.
The model's architecture is designed to handle complex development workflows, offering a robust alternative for teams looking to integrate advanced AI into their software engineering pipelines. By topping global coding and reasoning leaderboards immediately upon release, GLM-4.7 signals a shift in the competitive landscape for AI-driven development tools.
- Release Date: 2025-12-01
- Status: Open Weights
- Category: Coding & Reasoning
Key Features & Architecture
The GLM-4.7 architecture utilizes a Mixture of Experts (MoE) structure to optimize inference speed while maintaining high accuracy. It includes a specialized GLM-4.7 Flash variant, which is optimized for low-latency tasks and cost-sensitive applications. This dual-tier approach ensures users can choose between maximum performance or budget-friendly efficiency based on their specific project requirements.
The model supports a massive context window, allowing it to ingest entire codebases or extensive documentation in a single prompt. Additionally, it features multimodal capabilities, enabling it to process code screenshots and generate text explanations seamlessly. These technical specifications make it highly versatile for modern development environments where context retention is critical.
- Architecture: MoE (Mixture of Experts)
- Flash Variant: Low-latency optimization
- Context Window: 256k tokens
- Multimodal: Code image processing
Performance & Benchmarks
In terms of performance, GLM-4.7 has topped global coding and reasoning leaderboards upon release. It achieved a HumanEval score of 92.5%, surpassing previous industry standards for code generation tasks. On the MMLU benchmark, it scored 89.1%, indicating strong general reasoning capabilities across diverse domains.
Furthermore, it demonstrated a 76.8% pass rate on the SWE-bench Verified leaderboard, proving its ability to resolve real-world software issues autonomously. These concrete numbers validate its claim as a top-tier coding model, outperforming many closed-source alternatives in specific technical domains and establishing Zhipu AI as a serious contender in the global AI race.