AI News

Zhipu GLM-5.5 Expected to Launch by End of Month: Parameter Scale of 2.5 Trillion to 3.5 Trillion, Performance Rivaling Fable

Phoenix TechnologyUpdated Nov 16 18:14

AI summary

Zhipu's GLM-5.5 large model is expected to be released between late October and early November. An expert exchange summary circulating online states that its actual parameter scale is in the range of 2.5 trillion to 3.5 trillion, rather than the rumored 1.6 trillion. If it reaches more than 3 trillion, it will surpass the 2.8 trillion parameters of the Kimi K3 series and become the largest domestic AI large model. Sources say that after release, its performance can reach the flagship level of Fable, with the goal of providing large-scale service on the first day of release.

Body · AI translation

China's Largest and Most Powerful Large Model GLM-5.5 Expected to Launch by End of Month: Up to 3.5 Trillion Parameters, Performance Rivaling Fable

Kuai Technology Download the App for Exclusive Early Access October 21, 2026, 19:47:25 From Henan Updated October 21, 2026, 20:03:32

Kuai Technology, October 21 — In recent days, two domestic large models have already been released, DeepSeek V4.1 Flash-1017 and MiniMx M3.1 Flash, but no flagship-level large model has yet been launched. Among them, Zhipu's GLM-5.5 is said to reach the 3.5 trillion parameter level.

According to a Zhipu expert exchange summary circulating online, the GLM-5.5 large model that Zhipu plans to release at the end of October or early November is not the rumored 1.6 trillion parameters; its actual parameter scale is in the range of 2.5 trillion to 3.5 trillion.

By comparison, the largest domestic and open-source model in terms of parameter count is currently the Kimi K3 series, with 2.8 trillion parameters. If GLM-5.5 reaches more than 3 trillion, or even 3.5 trillion, it would be the strongest and largest domestic AI large model.

GLM-5.5's model architecture this time is relatively aggressive: it aims to train fast and infer fast while also lowering inference-side costs, with the goal of providing large-scale service on the very first day of release. Several cloud vendors have reportedly begun negotiations on a revenue-sharing model.

As for specific performance, it is said that problems in training have recently been resolved, and judging from the current situation, the next-generation model will reach the flagship level of Fable after release.

Domestic large models reaching Fable-level performance is something Zhipu founder Tang Jie previously confirmed. Musk previously believed that Chinese AI companies would not reach this level until next year, but Tang Jie said it would not take that long and that it could be achieved this year — clearly he has confidence in this.

This information cannot yet be fully confirmed, especially since the parameter count far exceeds the previously expected 1.6 trillion — equivalent to jumping directly from the current more than 740 billion to 3 trillion. A fourfold increase places very high demands on computing power and other resources.

As for Zhipu GLM-5.5's performance reaching the Fable level, judging from Tang Jie's previous statements, the credibility is very high. Zhipu's post-training capabilities are widely recognized as extremely strong; previously, with the more than 740 billion parameter GLM-5.2/5.3, it was already able to catch up to the frontier level.

"Special Statement: The content of the above works (including videos, pictures, and audio if any) is uploaded and published by users of 'Dafeng Hao,' a self-media platform under Phoenix Net. This platform only provides information storage space services.

Notice: The content above (including the videos, pictures, and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."

Source

Phoenix Technology