OpenAI Slashes Prices: Cheap GPT-5.6 Models Get Much Cheaper

OpenAI is reducing prices for its GPT-5.6 models Luna and Terra starting July 30, with Luna dropping by 80% and Terra by 20%. After the cut, Luna will cost $0.20 per million input tokens and $1.20 per million output tokens, while Terra will cost $2 and $12 respectively. Prices for the top-tier Sol model remain unchanged. According to OpenAI, Luna achieves performance similar to leading models from a year ago but at a much lower cost, estimating that a task costing $1 with those models would only cost about 6 cents with Luna, and Luna is also nearly nine times faster. The models are available through ChatGPT Work, Codex, and the OpenAI API. OpenAI attributes the price reduction to GPT-5.6 Sol itself making the infrastructure more efficient: the model independently optimized GPU software, cutting deployment costs by 20%, and improved token generation by over 15% via speculative decoding. The increasing price pressure in the AI market, particularly from cheap Chinese providers, likely also contributed. Microsoft is now openly promoting replacing OpenAI with its own cheaper MAI models. The price war could negatively impact the AI market if it reduces revenue growth for frontier labs, whose balance sheets are tied to extensive infrastructure investments.