Cerebras says its chips run a trillion-parameter AI model nearly 7 times faster than GPU clouds

Cerebras announced production support for Moonshot AI’s trillion-parameter Kimi K2.6 model.

Verified tests showed 981 output tokens per second, about 6.7x faster than the nearest GPU-based cloud.

The speed was also far ahead of the official Kimi endpoint, highlighting Cerebras’ wafer-scale inference approach.

The result aims to show that non-GPU systems can serve frontier models efficiently for enterprise workloads.