OpenAI cuts GPT-5.6 Luna prices by 80 percent as the AI price war heats up

OpenAI has cut the price of its cheapest GPT-5.6 model, Luna, by 80 percent, effective July 30. Luna now costs $0.20 per million input tokens and $1.20 per million output tokens. The mid-tier model, Terra, dropped 20 percent to $2 and $12. Sol, the most capable of the three, stays where it was.

A quick translation, in case those units mean nothing to you: a token is roughly three quarters of a word, so a million tokens is about 750,000 words, or a decent stack of books. Input is what you send the model, output is what it writes back. OpenAI says Luna now matches the quality of leading models from a year ago, and that a task which cost a dollar back then runs at about 6 cents on Luna today, roughly nine times faster. The new pricing applies across ChatGPT Work, Codex, and the OpenAI API.

Why now

OpenAI’s own explanation is that GPT-5.6 Sol made the company cheaper to run. According to OpenAI, the model optimized its own GPU software and cut deployment costs by 20 percent, and improved token generation by more than 15 percent using speculative decoding, a trick where a small fast model guesses ahead and a bigger model checks the guesses.

That is the flattering version. The less flattering one is competition. Chinese providers have been undercutting Western labs for months, and Microsoft is now openly marketing its own smaller MAI models as the cheap alternative to OpenAI. When your rivals are selling something close enough for a tenth of the price, you either explain why yours is worth more or you drop your price. OpenAI did the second. Worth keeping in mind: price wars are good for buyers and rough on sellers, and the frontier labs have enormous infrastructure bills tied to revenue that is now growing more slowly per token.

What this means for you: If you use ChatGPT through the normal subscription, nothing changes today. If you build anything on the API, or you run an automation that chews through a lot of text, this is real money. Summarising, tagging, drafting, sorting email, cleaning up data, all the boring high-volume work is now cheap enough that it is worth reconsidering tasks you shelved as too expensive six months ago. The honest caveat is that Luna is the small model. It is fine for volume work and noticeably weaker than Sol on anything that needs careful reasoning, so test before you switch a workflow over.

Sources

• OpenAI: Advancing the price-performance frontier with GPT-5.6
• The Decoder: OpenAI goes full China pricing mode