Media365
Beijing-based Zhipu AI, known internationally as Z.ai, on Monday evening claimed that its new GLM-4.7 model features improved coding abilities, narrowing the gap with larger US firms Google DeepMind and Anthropic in innovative coding agents.
Zhipu AI touted the high scores achieved by GLM-4.7 on various coding benchmarks that surpassed those of OpenAI’s GPT-5, which was released in August. GLM-4.7’s scores were just behind those of Google DeepMind’s Gemini Pro 3 and Anthropic’s Claude Opus 4.5.
On SWE-Bench, a benchmark that evaluates models on real-world software engineering issues, Anthropic broke the 70 per cent mark in May via its older Claude Opus 4 model, becoming the first closed-source company to do so. Zhipu AI’s GLM-4.7 achieved the same feat for the first time on Monday.
In a blog post on the same day, Zhipu AI said benchmarks were “only one way to evaluate performance”. It encouraged users to get a “feel” for how the GLM-4.7 model worked.
Hours later, Shanghai-based MiniMax introduced its M2.1 model, which also featured marked improvements in coding abilities on internal tests, while similarly lagging behind Anthropic’s Claude Opus 4.5.
