OpenAI just took a knife to its own API pricing, and the model most people actually use got cut deepest. GPT-5.6 Luna, the fast, cheap tier built for high-volume, routine work, dropped 80% in price: input tokens now run 20 cents per million, down from a dollar, output fell from $6 to $1.20. The mid-tier Terra got a smaller haircut, about 20%, landing at $2 per million input tokens and $12 for output. The flagship Sol kept its $5-and-$30 price tag untouched, but the tradeoff is speed: OpenAI says the API now runs 2.5 times faster.
The official line is a mission to make "advanced intelligence more abundant, affordable, and useful." The less flattering read is that enterprise AI bills have started to scare customers. Uber reportedly burned through its entire 2026 AI budget by the end of Q1 after engineers leaned hard on Claude Code. Microsoft is said to have paused Claude Code licenses altogether once spending blew past the annual plan within months. Survey data suggests nearly a third of companies now pour more than a quarter of their total cloud budget into AI alone.
None of this happens in isolation. Days earlier, OpenAI rolled out free access to its top models for tens of thousands of scientists, starting with 10,000 researchers and scaling to 100,000 by 2027 as part of a $250 million commitment. Put the two moves together and a pattern emerges: pull as many users as possible into the ecosystem while rivals like Gemini keep undercutting on price.
The harder question is how long OpenAI can keep API prices this low while it's also on the hook for hundreds of billions in compute spending over the next few years. Somebody in this race eventually eats the difference.
