VDraft takes first place globally in Google’s inference optimization competition, 'Fast Gemma Challenge'

Bidraft took first place worldwide on the revalidation benchmark at the Fast Gemma Challenge, co-hosted by Google’s Gemma team and Hugging Face.

Its autonomous AI agent, Bidraft-Darwin, delivered 510.58 tokens per second and a PPL of 2.3929, showing strong speed and quality together.

The result is seen as a benchmark case proving that inference optimization can make AI run faster and more efficiently on the same hardware.

It could help cut GPU usage and operating costs, with direct impact on AI service efficiency and infrastructure competitiveness.