OpenAI cuts GPT-5.6 prices by up to 80% in ChatGPT Work and API

OpenAI lowered the price of using two GPT-5.6 models on July 30, cutting GPT-5.6 Luna by 80 percent and GPT-5.6 Terra by 20 percent. The change applies across ChatGPT Work, Codex and the API. It is a meaningful reduction in model costs, but not a discount on the monthly price of ChatGPT Plus, Pro, Business or Enterprise subscriptions.

For API customers, Luna now costs $0.20 per million input tokens and $1.20 per million output tokens, down from $1 and $6. Terra has moved from $2.50 to $2 per million input tokens and from $15 to $12 per million output tokens. GPT-5.6 Sol pricing is unchanged. OpenAI says its subscription prices and quota budgets also remain the same, while Terra and Luna now consume fewer credits in ChatGPT Work and Codex. Eligible users can therefore do more work before reaching the same allowance.

Access depends on the plan. Free and Go users can use Terra, while Plus, Pro, Business and Enterprise customers can select both Terra and Luna. The lower underlying prices affect more than developers making API calls. People using the models for research, writing, analysis or coding may get more value from an existing subscription even though the amount charged each month has not changed.

OpenAI attributes the reduction to improvements in inference efficiency. The company says work on production kernels lowered end-to-end serving costs for Sol by 20 percent, while separate experiments improved token-generation efficiency by more than 15 percent. Those are company-reported figures, but they illustrate why model pricing can fall without releasing a new generation of hardware or reducing capability. OpenAI has also introduced a faster API mode for Sol, offering up to 2.5 times the speed at twice the standard price.

For developers and businesses, the new rates can materially change the economics of high-volume assistants, coding agents and automated analysis. An 80 percent reduction gives Luna more room to compete for workloads where cost previously outweighed quality. Terra becomes cheaper too, although the smaller cut may matter less for low-volume users.

The broader signal is that frontier-model pricing is becoming more dynamic. Efficiency gains are moving into customer prices quickly, while providers compete on cost as well as intelligence and speed. Buyers should still test total cost per completed task: a cheaper token is useful only if the model produces reliable results without requiring substantially more tokens, retries or human review.