GPT-5.6 makes the cheapest model 80% cheaper
OpenAI has cut the price of GPT-5.6 Luna while moving its faster API tier into a new Fast mode. The relevant change is not a benchmark: it is a different division of labor inside an agent workflow.
Luna now costs $0.20 per million input tokens and $1.20 per million output tokens. OpenAI says the cut makes high-volume tasks such as document analysis, classification, testing and routine implementation practical at a much wider scale.
The useful split is operational: use Sol for ambiguity and planning, then hand well-specified implementation and evaluation to Luna. Fast mode speeds up Sol up to 2.5×, but costs twice the Standard rate.
A cheaper worker, not a cheaper answer
OpenAI's July 30 announcement cuts GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%. The new API list prices are $0.20 and $1.20 per million input and output tokens for Luna; Terra is $2 and $12.
That moves the product question away from a single model choice. OpenAI's example is a workflow in which Sol resolves uncertainty and defines a plan, while Luna implements specified changes, runs tests and evaluates the result. The company says Luna can use tools and complete multi-step workflows, so the lower tier is intended to be an agent worker rather than a text-only fallback.
Fast mode puts a price on latency
The same release replaces Priority Processing with Fast mode for GPT-5.6 Sol. It offers up to 2.5 times the Standard speed at twice the price, without changing the model's intelligence; existing API requests tagged priority are mapped to Fast mode.
For teams, that makes latency a separate purchasing decision. A customer-facing or time-sensitive Sol step can pay for Fast mode, while the bulk of an asynchronous pipeline can use Luna. OpenAI says subscription prices and quota budgets do not change, though Terra and Luna consume fewer credits in Codex and ChatGPT Work.
The headline number is the 80% cut. The more consequential change is that a production workflow can explicitly spend its expensive reasoning on the small part of the job that is still uncertain.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
This is the last story in the July 31, 2026 issue.
Go to the issue archive