OpenAI cuts GPT-5.6 Luna prices by 80% and adds a faster paid API mode

OpenAI just made GPT-5.6 dramatically cheaper, and the shape of the cut says a lot about where the pressure is coming from.

What changed on the price list

On July 30, 2026, OpenAI cut API prices across the GPT-5.6 line: Luna now costs 80% less and Terra 20% less, while flagship Sol pricing stays unchanged.

The same update introduced a new Fast mode in the API, delivering up to 2.5x faster speeds than standard processing at twice the price. It replaces the old Priority Processing tier, giving latency-sensitive workloads a cleaner way to pay for speed.

So the menu now runs in both directions: much cheaper at the volume end, faster for a premium at the speed end, flagship untouched in the middle.

Why now

The context matters. Enterprise customers have grown cost-sensitive, and open Chinese models like Kimi K3 keep undercutting frontier pricing. A dramatically cheaper Luna is OpenAI’s answer for high-volume workloads: classification, extraction, and agents that make thousands of calls a day.

Those are exactly the workloads where open-weight alternatives have been winning on cost alone. An 80% cut moves the comparison from “an order of magnitude cheaper elsewhere” to a conversation OpenAI can stay in.

If you build on the API, your unit economics just changed overnight. Pipelines that were priced out of a frontier model last week may now clear the bar, which is worth a fresh look at anything you previously routed to a smaller model purely on cost.

Check the deprecation list

One housekeeping note from the same release deserves attention: the Evals platform, Agent Builder and reusable prompts are all headed for deprecation. If any of those sit in your production stack, check the migration timelines now rather than when they stop working.

The cheaper tokens are the headline. The quiet products being sunset in the same update are the part that can actually break your build.