At the 2026 semi-annual performance communication meeting held on the evening of August 31, some analysts raised questions about the market's “only post-training” intelligence and asked why the company had recently been relatively restrained in terms of parameter scale. Tang Jie, founder of Smart Spectrum, said that the next generation model will still expand the scale of the base, and at the same time control activation parameters to avoid a decrease in inference speed and a rise in costs. At the level of computing power, Smart Spectrum revealed that large-scale inference on 100,000-class domestic chips has been achieved, and the inference cost per token has decreased by 80% compared to the beginning of the year. GLM‑5.3‑Flash is SmartSpectrum's first model to fully rely on domestic chip clusters to provide services under ultra-large-scale real traffic. The company said that compared to the initial baseline of the same hardware, the model improved end-to-end service performance by 3 times. Tang Jie summed up one direction of future GLM-6.0 as self-evolution. “This is a big challenge. The biggest problem is that this model can judge for itself when to stop and when to correct itself. This is the biggest problem, rather than simply training the model. In future models, this is the focus of research.”

At the 2026 semi-annual performance communication meeting held on the evening of August 31, some analysts raised questions about the market's “only post-training” intelligence and asked why the company had recently been relatively restrained in terms of parameter scale. Tang Jie, founder of Smart Spectrum, said that the next generation model will still expand the scale of the base, and at the same time control activation parameters to avoid a decrease in inference speed and a rise in costs. At the level of computing power, Smart Spectrum revealed that large-scale inference on 100,000-class domestic chips has been achieved, and the inference cost per token has decreased by 80% compared to the beginning of the year. GLM‑5.3‑Flash is SmartSpectrum's first model to fully rely on domestic chip clusters to provide services under ultra-large-scale real traffic. The company said that compared to the initial baseline of the same hardware, the model improved end-to-end service performance by 3 times. Tang Jie summed up one direction of future GLM-6.0 as self-evolution. “This is a big challenge. The biggest problem is that this model can judge for itself when to stop and when to correct itself. This is the biggest problem, rather than simply training the model. In future models, this is the focus of research.”