Sol just got 14x faster and a third cheaper in eight days

Hook

The frontier model that needed US government clearance to ship in June is now the one OpenAI is discounting.

The news

Two OpenAI moves, eight days apart. On August 13 it previewed Ultrafast, a service tier that runs GPT-5.6 Sol, its top frontier model, at up to 750 tokens per second, roughly 14x standard speed, on the Cerebras hardware bet we covered in June; it's a limited preview for now, expanding "as capacity grows." Then on August 21 it cut Sol's price: input falls from $5 to $4 per million tokens (20%) and output from $30 to $20 (33%), through at least November 21, across the API, Codex credits, and eligible ChatGPT Work plans. Consumer plans stay put.

On your desk

Speed and price are the two dials that decide what a frontier model is actually for, and OpenAI just turned both at once. Any "too slow" or "too expensive" verdict you reached on Sol this summer expired this month. If a workflow, a vendor quote, or a build-vs-buy call got priced against July's numbers, the inference math underneath it moved by a third.

What you should do next

Rerun the cost math on anything you shelved for being too slow or too pricey, and if a vendor passes AI costs through to you, ask when the Sol cut shows up in your bill.