OpenAI released GPT-6.1 Terra on October 26, 2026. Standard API pricing is $1 per million input tokens and $5 per million output tokens—exactly half the cost of GPT-6.1 Sol and GPT-5.6 Terra. OpenAI says it's built for visual reasoning, computer use, and clear writing: the tasks that fill a normal workday.
Terra is faster than the pricier models for tasks like processing invoices, interpreting technical drawings, filling out online forms, and editing videos. Independent tests show it costs about $0.40 per task at maximum effort, compared to $1.40 for the prior Terra model. That's the gap that makes automation affordable for high-volume, repeatable work.
What you can do with it now
Pick one task your team does 20 times a week: pulling data from customer emails into a spreadsheet, drafting the same quote with different numbers, writing follow-up messages after a call, checking invoices against purchase orders. Set up Terra to handle that task in a test channel for a week. Compare the cost, the time, and the number of fixes you had to make.
If you're already using ChatGPT, switch to Terra for the tasks that don't need the smartest model: summarizing meeting notes, drafting routine replies, filling standard forms. Save the pricier models for the hardest decisions: complex proposals, coding, financial projections. OpenAI's own guidance: Terra is "similar on vision and computer use, a bit worse on most other things."
What to keep in your own hands
Cheaper doesn't mean careless. Terra makes mistakes—OpenAI's own report on its computer-use test identified six errors in one week, including production-data errors that went undetected for four days. Set the rule now: every Terra output gets a human check before it goes to a customer, gets billed, or changes a record. And keep the login: if the price doubles or the quality drops, you need to be able to walk away in a day, not renegotiate for a month.