China's Zhipu shows open source becoming China's edge in AI, seeking DeepSeek-like breakout!

Well, well, well! We now have Zhipu, from China, coming up with its own AI model, and quite similar to DeepSeek!

GLM-4.7 is said to be Zhipu flagship foundation model. It is reportedly featuring stronger encoding, intelligent agent, and general conversation capabilities. An earlier version, GLM-4.6, was earmarked as the world's 100B-level visual reasoning model with the best performance, natively supporting tool invocation.

Update: Today, Zhipu AI has released GLM-5 model in Hong Kong.

GLM-4.7 series is Zhipu's latest flagship model, enhancing coding capabilities, long-term task planning, and tool collaboration for Agentic Coding scenarios. It has achieved outstanding performance among open-source models in current leaderboards on multiple public benchmarks. GLM-4.7 reportedly improves agentic coding through better planning and tool use, greatly speeding up complex tasks.

General capabilities have been improved, responses are more concise and natural, and writing is more immersive. When performing complex agent tasks, instruction compliance is stronger during tool invocations, and the aesthetics of Artifacts and Agentic Coding front-ends and the efficiency of long-term task completion are further improved.

GLM-4.7 achieves significant breakthroughs in three dimensions: programming, reasoning, and intelligent agents:

Stronger programming capabilities: Significantly improved model performance in multilingual coding and on-device intelligent agents; GLM-4.7 can now implement a "think before you act" mechanism in programming frameworks such as Claude Code, Kilo Code, TRAE, Cline, and Roo Code, exhibiting more stable performance on complex tasks.

Improved front-end aesthetics: GLM-4.7 shows significant progress in front-end generation quality, capable of generating more visually appealing web pages, PPTs, and posters.

Stronger tool invocation capabilities: GLM-4.7 improves tool invocation capabilities, achieving a score of 67 in the BrowseComp web task evaluation; and a state-of-the-art (SOTA) score of 84.7 in the τ²-Bench interactive tool invocation evaluation, surpassing Claude Sonnet 4.5.

Improved reasoning capabilities: Significantly improved mathematical and reasoning abilities, achieving a score of 42.8% in the HLE ("Humanity's Last Test") benchmark, an improvement over GLM-4.6. 41%, surpassing GPT-5.1

General capabilities enhanced: GLM-4.7 dialogues are more concise, intelligent, and human-like; writing and role-playing are more eloquent and immersive.

Another Chinese startup?
This is yet another example of one more Chinese AI startup breaking into the USA, after DeepSeek. It is also said to be pulling in real coding demand from US developers. 

China AI coding demand has since exploded. As per Zhipu AI GLM coding agent, "Our online traffic has increased five-fold. This current subscription limit is a protective measure we have implemented to ensure a stable and high-quality experience for our existing users."

User base of Zhipu GLM Coding Plan is primarily concentrated in USA and China, followed by countries, such as India, Japan, Brazil and UK. The model is said to be very good, and gaining in popularity. It is very good on agents. This also highlights how open source is becoming China's edge in AI.

Like DeepSeek, Zhipu is also using open models as a distribution strategy as they are cheaper to run, easier to customize, and harder to be cut off from.

About seven of the top 10 AI models today are Chinese models. They are coding across both open and closed source. China is simply catching up! GLM 4.7 is the GPU model that is now in the news. Chinese companies are moving very quickly across the leaderboard. That's also a lesson for the USA and other countries to notice, absorb, and replicate.

DeepSeek shocked world, remember?
DeepSeek had shocked the entire world back in early 2025. Zhipu has drawn milder reactions so far. In fact, a lot of folks had tried to write off DeepSeek, saying, it was just a one-off! Until now! Even NVIDIA could feel threatened by the rise of Chinese open source AI models.

There have been several, and significant breadcrumbs from the various Chinese AI labs and companies that have showed they are making real progress. 

There is also the hardware side of things, and that can be interesting in the year ahead. We are seeing the progress that Huawei has been making. We are now seeing some new upstarts across China that have recently gone public, such as Zhipu.

Investors should keep their eyes on this spot, so that they don't get surprised in another DeepSeek-like moment!