OpenAI's new mid-range model costs half as much, until your prompt passes 272,000 tokens
On 26th Oct'26, OpenAI released GPT-6.1 Terra, its mid-range model for agent and coding work, at $1 per million input tokens and $5 per million output tokens. That is half the price of GPT-6.1 Sol and down from $2 and $12 for the GPT-5.6 Terra it succeeds. It is in the API from the Build tier and in Codex for paid ChatGPT plans. Any request with more than 272,000 input tokens is billed at twice the input rate and one and a half times the output rate for the whole request, and the default reasoning setting is medium, where higher settings spend more tokens on each task.
What it actually means
The halving holds for the short agent steps most teams run. It stops holding for the agent that keeps its whole history in the prompt, because once a request passes 272,000 tokens you pay $2 per million in, the same as the old Terra, and $7.50 out on every token of that request. A million-token window invites exactly that habit.
Before you switch, count how many of your own calls cross 272,000 tokens, and trim or summarise context so they don't. If you run on OpenAI's UAE inference residency, check before you move anything, because one launch write-up says Terra is not covered there yet.
Terra's half price only lasts while your prompt stays under 272,000 tokens.
272Kinput tokens where Terra's half price ends
Source: Developers Digest · Tuesday, 27 October 2026