OpenAI just made GPT-5.6 fourteen times faster and cut the price, weeks after launching it

OpenAI cannot stop touching the pricing dial on GPT-5.6, and it has only been out a matter of weeks.

Sol, the flagship model in the GPT-5.6 family, just got cheaper again. Input tokens dropped from $5 to $4 per million and output fell from $30 to $20 per million, a cut of more than 20 per cent, running through at least 21 November. It is the second GPT-5.6 price cut in under a month, after OpenAI trimmed pricing on the Terra and Luna models back in late July. This one covers the pay as you go API, Codex credits and eligible ChatGPT Work plans.

That comes just over a week after OpenAI previewed something arguably more interesting: an Ultrafast mode for Sol built on Cerebras hardware, running up to 14 times faster than the standard version and hitting speeds of roughly 750 output tokens per second. OpenAI has not published Ultrafast pricing yet, so treat that part as a preview rather than something you can switch on today.

None of this is happening in a vacuum. Google's Gemini 3.7 Flash launched the same week at roughly half the price of its previous generation with stronger agent performance, and Anthropic has reportedly been picking off enterprise accounts on cost. When three labs are all racing on speed and price in the same fortnight, that is not a coincidence, that is a proper price war.

For anyone actually building on these models, the maths keeps shifting under your feet, which is exactly the problem and exactly the opportunity depending on how you look at it. Locking into one provider right now feels like a fast way to leave money on the table a month later.

For everyone else, the upshot is simple. Frontier AI is getting faster and cheaper at the same time, and for once the labs racing each other are the ones eating the cost, not you. Enjoy it while it lasts, because price wars in tech have a habit of ending the moment the market consolidates.