MANY MINDEDOpenAI drops Luna token prices to a fifth of the old rate
OpenAI cut the price of GPT-5.6 Luna by 80% and Terra by 20%, effective July 30. Luna now runs $0.20 per million input tokens and $1.20 per million output tokens; Terra moves to $2 and $12. Sol, the top model, is unchanged. All three stay reachable through ChatGPT Work, Codex, and the OpenAI API.
OpenAI's framing is that Luna matches what the leading models of a year ago could do, and that a task costing a dollar on those models now runs about six cents on Luna, nearly nine times faster. The company attributes the cut to its own infrastructure getting cheaper: it says GPT-5.6 Sol optimised its GPU software, trimming deployment costs by 20%, and that speculative decoding lifted token generation by more than 15%. Those are OpenAI's claims about its own stack, unmeasured from outside.
Competition is doing at least as much work as engineering here. Low-cost Chinese providers have been pushing near-frontier quality far below Western list prices, and Microsoft is now openly marketing its MAI models as the cheaper option. The attribution matters less than the direction: a capability that cost a dollar a year ago costs cents today. That is what it looks like when something stops being scarce — not a discount, a repricing.
The risk sits on the other side of the ledger. Frontier labs carry balance sheets tied to enormous infrastructure commitments, and a price war that slows revenue growth could squeeze the very buildout that made these cuts affordable. Watch whether Sol's price holds, and whether the cheapest tier stays this cheap once the competitive pressure eases.
Source: The Decoder