What actually shipped
On July 9, 2026 OpenAI made GPT-5.6 publicly available in three named variants, and the pricing is the story. Sol is the most powerful tier at 5 dollars per million input tokens and 30 dollars per million output tokens. Terra is the everyday tier, priced at 2.50 dollars per million input tokens and 15 dollars per million output tokens, with performance comparable to the previous generation GPT-5.5. Luna is the lowest-cost tier at 1 dollar per million input tokens and 6 dollars per million output tokens.
The rollout followed a controlled path. OpenAI first released these models to approved partners only in late June, following a US administration AI cybersecurity order. The US Department of Commerce Center for AI Standards and Innovation, known as CAISI, ran additional testing before clearing the broader public rollout, which was announced July 8 and went live July 9.
Why the price cut matters more than the benchmark
The news is not the benchmark, it is the repricing. Terra delivers performance comparable to GPT-5.5 at 2.50 dollars per million input tokens, which is 50 percent below what the same class of model cost a generation ago. For the large share of workloads that never needed a flagship, classification, extraction, summarisation, routine drafting, the everyday unit price has just halved.
Luna pushes the logic further. At 1 dollar per million input tokens and 6 dollars per million output tokens, it sets a new floor for high-volume, low-complexity work. The practical effect is that the cost of running AI at scale has moved, even where the model quality has not changed much for your particular task.
What to do about it
The owner move is to reprice the AI budget and route work by task tier, not to celebrate a new leaderboard position. Start by separating your calls into three buckets: work that genuinely needs the flagship, everyday work that Terra now covers at half the old price, and high-volume work that Luna can absorb at the floor. Most organisations discover the first bucket is far smaller than their current bill assumes.
Then reset the budget to the new prices rather than the old ones. If your everyday AI spend was sized against last year's rates, roughly half of it is now recoverable or redeployable. Defaulting every call to the top model is no longer a neutral choice, it is a line item you are choosing to keep paying.
Read next: Washington Now Sets When Frontier AI Reaches You | The US Can Now Delay the AI Model You Rely On
