OpenAI is putting its newest frontier model on AI's fastest chip

Hook

Open chat.cerebras.ai and ask it anything. The whole answer lands before you've finished reading your own question, and that is the speed OpenAI just bet its newest frontier model on.

The news

Today OpenAI previewed GPT-5.6 in three sizes: Sol, the top-tier frontier model, plus a cheaper Terra and a faster Luna. The line operators should circle: Sol launches on Cerebras hardware in July at up to 750 tokens per second, roughly 560 words a second, which OpenAI calls "frontier intelligence at unprecedented speed." It opens as a limited preview, gated to about 20 companies after OpenAI cleared the release with the US government. Source.

On your desk

Speed is not a spec here, it is a feel. Most AI you use trickles out at 50 to 150 tokens a second, so you sit and watch it think. Cerebras runs open models, including OpenAI's gpt-oss, at around 1,900 tokens a second, roughly 2.7x the next-fastest host and more than 10x a normal chat.

When the wait disappears, the tool stops feeling like a slow oracle and starts feeling like a fast colleague. That changes what you reach for it to do. OpenAI sees it too: in January it signed a multi-year deal worth more than $10 billion to run its inference on this same hardware, and now its newest frontier model is headed there.

What you should do next

Spend two minutes at chat.cerebras.ai before your next meeting and ask it something real: a tricky reply, a messy summary, a gut-check on a call you're about to make. Watch the answer just appear, and notice how hard the old pace is to go back to. Tell me what you ask it.