GPT-5.6 pricing just had the kind of change that should make anyone running an agent reopen a spreadsheet. Luna is now one fifth of its old price. Terra is cheaper too. Your ChatGPT subscription, however, did not suddenly get an 80% discount.
OpenAI cut GPT-5.6 Luna API prices by 80% and Terra prices by 20% on July 30. Luna now costs $0.20 per million input tokens and $1.20 per million output tokens. Terra costs $2 for input and $12 for output.
The company also introduced Fast mode for GPT-5.6 Sol. It can run up to 2.5 times faster than standard processing at twice the API price. Sol’s base price did not change.
The new GPT-5.6 pricing
API prices per one million tokens, effective July 30, 2026.
GPT-5.6 pricing changed more than the sticker
The Luna cut is the important one. It turns a model that can use tools and complete multi-step work into a realistic option for jobs where hundreds or thousands of runs would previously have felt wasteful.
Think classification, document cleanup, first-pass code changes, test execution, or background agents that handle well-defined tasks. Those jobs do not need the smartest available model every time. They need a model that is good enough, consistent, and cheap enough to call repeatedly.
How deep were the cuts?
The bar shows the share of the previous price that remains.
OpenAI says Luna now delivers performance comparable to models that were frontier-class a year ago at roughly six cents on the dollar per task and close to nine times the speed. Those are vendor measurements. Treat them as a reason to test, not as a result you have already earned.
ChatGPT plans did not become 80% cheaper
This is the detail most likely to get lost in a viral headline. OpenAI kept ChatGPT and Codex subscription prices unchanged. Quota budgets also stay the same.
Paid users should see Terra and Luna consume fewer credits. Free and Go users can access Terra, while Plus, Pro, Business, and Enterprise users can choose Terra and Luna. The savings arrive as more usage inside the same plan, not a lower monthly bill.
If you only use ChatGPT in a browser, do not multiply your monthly fee by 0.2 and call it the new price. GPT-5.6 pricing has two layers: API token rates and subscription credit consumption.
Fast mode is a latency tax you can choose
Fast mode replaces Priority Processing in the API and lines up with the /fast option in Codex. OpenAI says Sol can respond up to 2.5 times faster at twice the price, with no change in intelligence.
That trade is easy to understand and easy to misuse. Paying double makes sense when a person is waiting, a production incident is burning time, or a model response blocks the next step in an expensive workflow. It makes little sense for overnight jobs, queued evaluations, or background code review.
A practical GPT-5.6 routing map
Choose by cost of error, volume, and urgency.
High-volume, clear instructions, routine tool use, and repeatable background work.
Everyday knowledge work where quality and cost both matter.
Hard uncertainty, high-stakes planning, or work where waiting costs more than tokens.
The best way to use the cut
Do not move your whole application to Luna because one chart looks good. Pick one expensive workflow and run a controlled comparison.
- Collect 20 real tasks, including the ugly edge cases.
- Run the same inputs through your current model and the cheaper candidate.
- Score correctness, tool completion, latency, and total tokens.
- Route only the tasks where the cheaper model meets your quality floor.
The cost per token is not the cost per successful task. A model that needs three retries can erase a discount. A model that completes the job with fewer output tokens can beat a lower-priced rival.
My read: Luna is now the model to challenge your default
The new GPT-5.6 pricing does not make model choice automatic. It does make an old habit harder to defend: sending every task to a flagship because changing the router feels annoying.
I would keep Sol for the uncertain part of a job, use Terra where judgment still matters, and make Luna prove itself on the repeatable work. If Luna passes your own evaluation, the 80% cut is real money. If it fails, the cheap price is trivia.
Go deeper
- Read OpenAI’s official GPT-5.6 pricing announcement.
- See why GPT-5.6 Sol’s tool loop matters more than one score.
- Use our real-work model comparison method.
- Read why power users adopt a multi-model workflow.
Updated August 4, 2026 with OpenAI’s July 30 rates. API prices, subscription access, and usage policies can change, so check the official pricing page before setting a production budget.
