Luna's price cut by eighty per cent, Sol's left where it was
On 30 July 2026 OpenAI cut the price of GPT-5.6 Luna by 80 per cent and of GPT-5.6 Terra by 20, leaving the flagship Sol unchanged. A million input tokens of Luna fell from a dollar to twenty cents and output from six dollars to 1.20; Terra went from 2.50 to 2 and from 15 to 12.
Why it matters
The cut reached only the cheap and middle tiers, which makes it an answer to pressure from below rather than a general fall in prices. Reuters ties it to closer scrutiny of enterprise AI bills and to competition from Chinese open-weight models. OpenAI's own explanation is worth its own note: part of the saving came from Sol working on the infrastructure that serves it — a self-improvement claim that this time carries a price tag.
The announcement came on 30 July 2026. Prices per million tokens after the cut: Luna at 20 cents for input and 1.20 dollars for output, down from a dollar and six; Terra at 2 and 12, down from 2.50 and 15. Sol's price was left unchanged. The cut also flows into how Luna and Terra usage counts against paid subscription limits in Codex and ChatGPT Work. The same day the API gained Fast mode, replacing priority processing: for Sol it gives up to 2.5 times the speed at twice the rate with no change in intelligence, and requests marked priority move to it automatically. The company names its own reason: efficiency gains from GPT-5.6, including the model's ability to improve code and optimise performance during internal development. By its account, in a human-led process Sol autonomously rewrote and optimised production kernels, designed and ran hundreds of experiments and watched over training; the kernel work brought about a 20 per cent reduction in total cost of serving, and the experiments better than a 15 per cent gain in token generation efficiency. These numbers come from OpenAI and are not independently verified. Reuters supplies the market side: the cut raises pressure on Anthropic, whose models lead enterprise and developer use but sit at the costlier end — its mid-tier Claude Sonnet 4.6 costs 3 dollars per million input tokens and 15 per million output, above Terra. Both companies are meanwhile under pressure from Chinese open-weight models such as Z.ai's GLM-5.2, which nearly match them at a lower price.