the surge of domestic large models is bringing them closer to the top American closed-source AI models to a point where it's hard to tell the difference. According to Qianzhi, on July 20th, after Qwen and Kimi showcased their strength, it's now Zhizhu's turn. The rumored GLM-5.5 is described as a historic upgrade. This time, it's not media speculation but Zhizhu's founder Tang Jie who has stepped into the spotlight.

A netizen mentioned in comments that both Qwen and Kimi have undergone a historic evolution, and casually asked if there was still hope for GLM. Tang Jie's response immediately raised expectations—he used the term "epic plus." This means that according to Zhizhu's own judgment, the improvement in this generation is even greater than what the netizen mentioned.

image.png

Of course, specific numbers are not available yet, but Tang Jie clearly hopes the outside world maintains confidence in GLM. His confidence comes from the achievements of the previous generation: before Kimi K3 was released, GLM-5.2 was the first domestic large model to match the performance level of Opus in AI programming ability. More notably, its scale—GLM-5.2 has only around 74 billion parameters, roughly a quarter of Kimi K3 and about one-fifth or even less than top American AI models—yet it managed to bring its capabilities up to that level through post-training, serving as a prime example of achieving great capabilities with small parameters.

The new model referred to by Tang Jie as "epic plus" is likely the next generation, GLM-5.5. Previously, there were rumors about GLM-5.3, but now it seems GLM-5.5 is more likely, although it's also possible that it might be named GLM-6 at launch. One thing that can be confirmed is that the parameter count of the next generation of large models will at least exceed 1 trillion. Considering that Zhizhu has previously used DeepSeek's open-source base model, it's speculated that GLM-5.5 may reach 1.6 trillion parameters, comparable to V4Pro; combined with Zhizhu's well-known strong post-training capabilities, its performance leap is definitely worth betting on, possibly marking another domestic large model comparable to Fable5 in scale.

However, in the waiting list of the domestic lineup, the most anxious moment right now is the official release of DeepSeek V4. At the time the report was published, it was the last moment of mid-July—the midnight hour was hanging overhead. Whether this little whale would suddenly unveil a big move in the middle of the night or once again delay the release is uncertain. When Zhizhu ignited the topic with an "epic plus," it's clear that the relay race of domestic large models is far from over.