Powered by CerebrasLight-speed.
Light-speed.
Literally.
Every reply streams at roughly a thousand tokens a second on the fastest inference humans have built. The answer's done before you've leaned back.
On Cerebras
The fastest inference
on Earth.
Most chatbots make you watch a cursor blink. Krex doesn't blink. Whole essays in a heartbeat, long answers before the kettle boils.
~1,000
tokens / second
<2.5s
to first token
∞
usage, in beta
Benchmarks
Faster than what you already pay for.
Measured against the models you know. Shorter bars are faster.
Streaming speed — tokens / second
Reasoning — time to a correct multi-step answer
Illustrative figures based on typical streaming throughput. Swap in your own measured numbers before launch.
~1,000
tokens streamed per second
<2.5s
to first token
∞
usage while in beta
Choose how to get in
Invite-only. No Google, no Meta, no email.
krex.space · new chat
Just keep swimming.Ask anything — watch it stream.
+✦ Auto
idleSealed end-to-end · ~1,000 tok/s on Cerebras