LIVE
Loading prices...
View All

Zhipu AI Launches GLM-4.7, Claims Parity with GPT-5.1 and Claude

Zhipu AI announces GLM-4 7, claiming performance parity with OpenAI's GPT model in a presentation slide.

Zhipu AI just dropped a major new open-weights model, GLM-4.7, and they’re making some bold claims. Released on December 22, 2025, this model is designed for autonomous agents and reportedly matches the performance of top-tier proprietary models like OpenAI’s GPT-5.1 and Claude Sonnet 4.5 (as referenced in performance claims by Zhipu AI), as claimed by Zhipu AI.

This is a significant move in the AI landscape, especially as other major players tighten their grips on closed ecosystems. For developers and companies looking for powerful, customizable AI without vendor lock-in, GLM-4.7 presents a compelling alternative. Its open-weights nature under an MIT license means local tech teams can potentially fine-tune and deploy it for specific regional needs.

The model introduces a key feature called ‘Preserved Thinking.’ This is built to stop context decay in complex, multi-step workflows. Think of an AI agent planning a detailed project or writing long code, this feature helps it remember its own reasoning chain from start to finish, which is crucial for reliability.

Also GLM-4.7 shows a massive 38% improvement over its predecessor, GLM-4.6, on the HLE benchmark. It also scored a 95.7 on the challenging AIME 25 benchmark. The focus is squarely on coding, tool use, and complex reasoning tasks.

For developers, the accessibility is a big deal. You can access it via Z.ai’s API, with a coding plan starting at just $3 per month. More importantly, you can download the full model from Hugging Face or ModelScope and run it locally. This is huge for projects with data privacy concerns or for those wanting to avoid ongoing API costs.

What do you think? Could an open model like GLM-4.7 change how AI is adopted in the tech scene? Let us know your thoughts in the comments.

Zhipu AI Launches GLM-4.7, Claims Parity with GPT-5.1 and Claude


Model: GLM-4.7
Parameters: 358.0 Billion
Context Window: 200K tokens
Max Output: 128K tokens
Benchmark (AIME 25): 95.7
Benchmark (HLE): 42%
Improvement over GLM-4.6: 38%
License: MIT
Release Date: December 22, 2025
Access: Z.ai API, Hugging Face, ModelScope
Local Runtime: vLLM, SGLang
Z.ai Coding Plan: Starts at $3/month

Founder of Pinoyscreencast, started using YouTube as a medium to disseminate Filipino-spoken technical tutorials. He decided to embark on reviews focusing on aordable gadgets. As he kept sharing more content, his subscriber base grew and shared how his videos influenced them in making a product purchase. Randolph is an experienced video content creator, he worked for a video production company for almost four years, from Shooting, Editing, Compositing and upto online content publishing.

1513 posts

Comments

Your contact info is private.

No comments yet. Be the first to share your thoughts!