OpenAI just cut the price of GPT 5.6 Luna by 80%. And the model itself is the reason they could.
The New Prices
GPT 5.6 ships in three sizes: Sol at the frontier, Terra for everyday work, and Luna as the fast, cheap one. On July 30, OpenAI dropped Luna's price by 80% and Terra's by 20%.
Luna now runs 20 cents per million input tokens and $1.20 per million output. Terra went to $2 in and $12 out. For comparison, Sol is still $5 in and $30 out. Luna is not a cheap model. It is a cheap frontier model.
Punching Above Its Price
On the Artificial Analysis Intelligence Index, Luna scores above Claude Opus 5, Gemini 3.6 Flash, and GLM 5.6 Max while sitting an entire order of magnitude to the left on cost. OpenAI's own claim goes further: on Agents' Last Exam, Luna beats Claude Fable 5 at an estimated cost per task nearly 99% lower.
The Model Made Itself Cheaper
This is the part worth stopping on. Inside a human-led process, GPT 5.6 Sol autonomously rewrote and optimized OpenAI's production kernels and ran hundreds of its own experiments on token generation. The kernel work cut the end-to-end cost of serving the model by up to 20%. The experiments pushed token generation efficiency up more than 15%.
Essentially, the model made itself cheaper to run, and you are getting the invoice.
Fast Mode Hits the API
There is a second change developers will hit immediately. Fast mode is now in the API, up to 2.5 times faster than standard on Sol, at double the price, with no change in intelligence. Requests already tagged priority switch over automatically.
If you route your workloads between Sol for the hard problems and Luna for the volume, your API bill just changed shape. The frontier did not get smarter this week. It got an order of magnitude cheaper.