GPT-6.1 Terra makes near-flagship computer use cheap, and puts office automation in reach

GPT-6.1 Terra is a faster, cheaper model aimed at everyday office work: drafting, reading technical drawings, filling out forms, editing video and operating desktop apps. API pricing is $1 per million input tokens and $5 per million output, which OpenAI says is half the price of both GPT-5.6 Terra and GPT-6.1 Sol. It is live for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API.

Terra is OpenAI’s efficiency line, below GPT-6.1 Sol (Sol launch) and GPT-6 Astra. OpenAI says Sol and Astra remain better for the most complex coding, while Terra more than doubles its predecessor on Terminal-Bench 4.0, solving 50.3% of tasks at max effort versus 21.5%. For how the family fits together, see our GPT-6 model guide.

OpenAI says Terra at medium effort scores 15 points higher than its predecessor on its computer-use benchmark at about a third of the cost, and matches GPT-6 Astra at a tenth of the price per token. When near-flagship ability is this cheap, the question shifts from whether a company can afford to automate a workflow to whether it should.

The most useful AI right now is the model cheap and reliable enough to do dull, high-volume work. OpenAI’s own example, reading incoming invoices and entering them into accounting software, is that kind of job. A model that does it well at this price will be deployed widely, and many people will get time back.

On OpenAI’s internal set of error-inducing prompts, Terra’s error rate at low effort fell from 53.6% to 16.7%, and to 4.3% at max effort. That’s a large improvement. It’s also an internal benchmark built and scored by the seller. For an agent clicking through real software, even a small error rate means wrong entries in real ledgers. Cheap automation multiplies mistakes as efficiently as output.

OpenAI also targets Anthropic directly, saying Terra at max effort is more than three times as token-efficient as Claude Opus 5.5 and about nine times cheaper per task on its computer-use test. Those are vendor numbers on vendor-chosen benchmarks, so independent testing should come first. The direction is clear, though: capable agents are getting commoditized fast. Our frontier model comparison found the real differences are increasingly price and access.

OpenAI says Terra meets the Critical cybersecurity threshold in its Preparedness Framework, keeps Astra’s safeguards and resists prompt injection well in its tests. But the cheap, fast, broadly available option built for operating computers now carries capability OpenAI itself rates Critical. Attackers get the price cut too, and the safeguards must hold at much higher volume. OpenAI’s specialist GPT-6 Cyber is gated tightly by comparison (our analysis).

A model marketed for the communication and computer tasks that fill the workday is aimed at work people are paid to do now. The gains are real, and companies should be honest about whether time saved goes to workers or to fewer of them. More in our future of work coverage and OpenAI coverage.