In partnership with
On July 30, OpenAI cut the price of its cheapest model by 80%.
GPT-5.6 Luna launched on July 9 at $1.00 per million input tokens. Twenty-one days later it costs $0.20. Output fell the same way, from $6.00 to $1.20. Terra, the mid tier, dropped 20% to $2.00 and $12.00. Sol, the flagship, did not move at all.
Price cuts are ordinary. This one matters because of how OpenAI says it found the savings.
An AI worked on the software that runs the AI
The day before the price change, OpenAI's engineering team published how it made GPT-5.6 cheaper to serve. Two changes stand out, and OpenAI credits its own flagship model, GPT-5.6 Sol, with both.
First, kernels — the low-level programs that do the actual math on a GPU. Fiddly, specialist work. OpenAI put Sol to work rewriting its production kernels inside Codex, using Triton and Gluon, two open-source GPU languages OpenAI maintains. That work, plus related improvements Sol found, cut end-to-end serving costs by 20%. Because subtly wrong math code is worse than none at all, OpenAI validated the output with FpSan, its open-source floating-point checker.
Second, speculative decoding — a standard trick where a small draft model guesses the next few tokens and the big model verifies them in a single pass instead of generating them one at a time. Sol ran hundreds of experiments on that draft model's design, then launched and supervised its training, stepping in on its own when hardware failed. Token-generation efficiency rose by more than 15%.
Neither number is 80%. Those gains created the margin. The size of the cut was a business decision on top of it.
Don’t Let Tax Season Cost You Year-Round
That pit in your stomach is trying to tell you something: Waiting until spring is costing you peace of mind.
When tax season feels like a crisis, it’s usually because the right financial information isn’t organized ahead of time. Deductions, education expenses, and important documents all become a last-minute scramble.
Listen to your gut. You can start preparing now.
BELAY’s experienced tax prep professionals help you stay organized year-round, so tax season becomes simpler, less stressful, and actually manageable.
Start with BELAY’s free Personal Tax Prep Checklist and take the first step toward a smoother tax season.
Don’t spend another spring stressing over paperwork. Get help now and leave the pit in your stomach behind for good.
The four words worth noticing
In the engineering post, the claim is bare: Sol autonomously rewrote and optimized our production kernels.
In the pricing post the next day — the one customers read — the same work happened within a human-led process. That qualifier appears twice there, and not once in the engineering post.
Both posts are OpenAI's. Neither is false. But when the company with every reason to sell our-AI-improves-itself is the one adding the human-in-the-loop caveat, the quieter version is usually the truer one.
It echoes the July 9 story that Sol post-trained the smaller Luna model. Real — and narrowed in public by OpenAI researcher Jason Liu, who noted that most of the configuration already existed and the actual job was adapting it and running it, saving perhaps two researchers two weeks. Significant. Not a machine inventing a recipe from nothing.
This is not the chip
Edition 292 covered Jalapeño, OpenAI's first custom inference chip, and noted that no price cut had been tied to it. Still true. This one came from software.
What it means for you
If you build on the OpenAI API, the math under your workflows changed and nobody sent a memo. Work that was marginal at a dollar per million input tokens — bulk extraction, classification, first drafts, the routine steps inside agent loops — now costs a fifth of that.
The move is not to switch everything to the cheap model. It is to take one workflow you already run and ask which steps ever needed the expensive one. Most were built when a single tier was the only sane default. That default is now three weeks old.
On the site version there's a calculator you can drag to your own monthly volume, plus the copy-paste prompt that walks you through that audit step by step. Both of OpenAI's posts are linked there too — the engineering writeup and the pricing announcement.
Free email without sacrificing your privacy
Gmail tracks you. Proton doesn’t. Get private email that puts your data — and your privacy — first.
— The AI Super Simplified team