Cerebras

Cerebras delivers the fastest public inference for large models (~2,600 tok/s) with 1M free tokens per day — ideal for rapid agent loops and big-model tasks that stay within short context windows.