According to CoinMeta, during its 2026 semi-annual performance communication meeting, Zhispu responded to market doubts regarding its approach of "only using post-training." Founder Tang Jie stated that the next-generation model GLM-6.0 will continue to expand the base scale and implement control over activation parameters to avoid a decline in inference speed and an increase in costs. Zhispu revealed that it has achieved large-scale inference using 100,000-level domestic chips, with the inference cost per token unit decreasing by 80% compared to the beginning of the year. GLM-5.3-Flash is Zhispu's first model that fully relies on a domestic chip cluster under ultra-large-scale real traffic, with performance improved by three times. Tang Jie pointed out that one direction for GLM-6.0 is self-evolution, emphasizing that the model needs to possess the ability to make self-judgments and corrections, which is a key focus for future research.