SmartPuzzle Token Inference Cost Down 80% Since the Beginning of the Year: 100k Domestic Chips Running Large Models

Voyager Beating AI News Flash: The Wisdom Spectrum has revealed that the company has successfully achieved large-scale low-cost inference of a 100,000-level domestic chip. The unit token inference cost has dropped by 80% compared to the beginning of the year.

This batch of domestic computing power has previously undergone real traffic stress tests. Prior to the release of GLM-5.3-Flash, Ox-Alpha anonymously launched OpenRouter and OpenCode, with a token invocation volume of 620 trillion. After the official release, all online traffic continues to be supported by 100,000 domestic chips.