According to CoinMeta, Zhipu’s new model is reported to have up to 3.5 trillion parameters and is expected to reach the Claude Fable level. A summary of an expert interview that has been circulating in investment communities states that Zhipu’s next-generation base model does not have 1.6 trillion parameters as previously rumored, but rather ranges from 2.5 to 3.5 trillion parameters. Zhipu’s glm-5.3 released in August this year still uses the previous generation base model, with a total of about 744 billion parameters. If the information is true, the scale of the new base model will be approximately 3.4 to 4.7 times that of the current one. The summary mentions that Zhipu adopted a more aggressive model architecture this time, and during the pre-training phase, there was an issue where loss did not decrease. The interviewees believe that this problem should have been resolved recently, and they expect the training to be completed soon. They predict that the new model will be able to reach the flagship level of Claude Fable. The new model also aims to solve the problem of being able to produce large quantities but not keeping up with demand after the release of large models. From the architectural design stage, Zhipu has set the inference speed and cost as strict constraints, with the goal of being able to operate on a large scale from the first day of release. Moreover, multiple cloud service providers are already in talks regarding revenue sharing.