BEIJING: As global expectations for AI shift from capability demonstrations to dependable deployment, Z.ai has introduced GLM-4.7 — a model engineered not merely to impress in benchmarks but to operate reliably within demanding real-world engineering environments. Representing a major step beyond GLM-4.6, the new model brings stronger execution stability, improved reasoning coherence and measurable performance leadership among open-source competitors.
Modern development environments increasingly require AI that can handle lengthy multi-step workflows while maintaining behavioral consistency, particularly when interacting with tools and external systems. GLM-4.7 is trained with these realities in mind, demonstrating improved resilience across extended coding processes, terminal-agent operations and multi-language programming conditions.
Performance testing confirms this positioning. GLM-4.7 scored 87.4 on τ²-Bench — the highest known score among open-source tool-use models — and 67.5 on BrowseComp. Meanwhile, across major coding benchmarks, it performs at or above Claude Sonnet 4.5, while significantly outperforming GLM-4.6.
One of the model’s defining strengths lies in reasoning control. GLM-4.7 is capable of sustaining logical consistency over multiple task interactions while dynamically scaling reasoning depth based on complexity. This significantly enhances predictability — a key requirement as AI systems become embedded within enterprise production pipelines.
Beyond coding, GLM-4.7 demonstrates greater maturity in UI and presentation generation. Its handling of layout structure, spacing, design hierarchy and styling coherence creates outputs requiring fewer manual revisions — a notable step toward closing the gap between AI conceptual drafting and production-ready content.
Importantly, its conversational fluency, writing tone and creativeness have also evolved, enabling broader deployment across communication, interactive applications and creative fields.
GLM-4.7 is fully integrated within BigModel.cn and Z.ai’s development ecosystem, with growing adoption across leading developer platforms including TRAE, Vercel, YouWare, Cerebras, OpenRouter and CodeBuddy — reflecting expanding trust and ecosystem adoption.
Complementing its technological advancements, Z.ai revealed plans to list on the Hong Kong Stock Exchange, aiming to become the world’s first publicly listed company whose core business is large-model AGI development. With revenue rising from RMB 57.4 million in 2022 to RMB 312.4 million in 2024 and continuing strong growth in 2025, the IPO marks a pivotal milestone both for the company and the broader AI industry.
GLM-4.7 therefore arrives not just as a technological upgrade but as a strategic signal — that open AI built in China is ready to compete at the highest global level while supporting real engineering, real businesses and real-world deployment.