Token reasoning cost 80% lower than at the beginning of the year: 100,000 nationally produced chips run large models

According to the 31 August news, the spectra revealed that the company had now achieved a scaled low-cost reasoning of 100,000 class of nationally produced chips, and that the unit cost of the Token reasoning was 80 per cent lower than at the beginning of the year. This amount of national production was previously measured by real flow pressure. OpenRouter and OpenCode, Token, reached 62 trillion before GLM-53-Flash was released anonymously; the total online traffic also continued to be carried by 100,000 national chips after the official release。