OpenAI just cut prices again on its GPT-5.6 family of AI models, and the cut is steep. Starting July 30, Luna, the cheapest and fastest model in the lineup, dropped in price by 80 percent. Terra, the mid-tier model, dropped by 20 percent, while Sol, the flagship model, kept its original price.
In plain terms, Luna now costs 20 cents to read one million words worth of text and $1.20 to write that much back. Terra costs 2 dollars to read and 12 dollars to write at that same scale. These prices apply whether a business is using ChatGPT directly, the coding tool Codex, or plugging into OpenAI's systems through its own software.
OpenAI says a task that would have cost a dollar to run through last year's best AI models now costs about six cents on Luna, and finishes nearly nine times faster. The company says this is possible partly because its own newest model was used to rewrite the software that runs OpenAI's computers, cutting operating costs by a fifth. It also sped up how quickly the model produces answers by more than 15 percent, using a technique where the AI drafts several guesses at once and checks them together instead of writing one word at a time.
This is happening because OpenAI is not the only option anymore. Chinese AI companies including DeepSeek and Alibaba have cut their own prices repeatedly through 2026, and some of their models remain cheaper than OpenAI's new rates even after this cut. At the same time, Microsoft, one of OpenAI's largest investors and its biggest distribution partner, has been pushing its own in-house models as cheaper replacements inside everyday tools like Word, Excel, and Outlook.
The timing matters. OpenAI is spending enormous sums building data centers and buying computing power, with commitments running into the hundreds of billions of dollars, while the company overall is still losing money. Cutting prices while burning cash looks risky, but it only pays off if enough businesses shift more of their work onto these models than they already do.
For any business that uses AI tools, directly or through software built on top of them, the immediate takeaway is simple: whatever job looked too expensive to automate a year ago is worth testing again now. The bigger lesson is that prices in this market are moving fast and often, and no single AI provider's pricing should be treated as fixed. Businesses that build their budgets, contracts, or workflows around one provider's current prices should expect more shifts like this one, and it is worth comparing options regularly rather than assuming today's rate holds next quarter.