Introducing GPT-6.1 Terra

OpenAI has released GPT-6.1 Terra, a faster, cheaper model aimed at the everyday work of the office: drafting, reading technical drawings, filling out forms, editing video and operating desktop apps. Standard API pricing is $1 per million input tokens and $5 per million output, which OpenAI says is 50 percent below both GPT-5.6 Terra and GPT-6.1 Sol. It is live today for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API.

The headline is cost. OpenAI says that on its computer-use benchmark, Terra at medium effort scores 15 points higher than its predecessor at about a third of the cost, and performs on par with the flagship GPT-6 Astra at one-tenth the price per token. When near-flagship ability gets this cheap, the question stops being whether a company can afford to automate a workflow and becomes whether it should.

There is also a safety marker that deserves attention. OpenAI says Terra meets the Critical threshold in cybersecurity under its Preparedness Framework, and carries over the protections built for GPT-6 Astra.

We are broadly positive on this launch. The most useful AI right now is not the model that wins a hard math contest; it is the one cheap and reliable enough to do the dull, high-volume work that eats people's days. OpenAI's own example, reading incoming invoices and entering them into accounting software, is exactly that kind of job. A model that can do it well at this price will get deployed, and a lot of people will get time back.

The factuality numbers are the part we find most encouraging, with a caveat. On OpenAI's internal set of error-inducing prompts, the error rate at low effort fell from 53.6 percent to 16.7 percent, and at max effort it dropped to 4.3 percent. That is a big improvement. It is also an internal benchmark, built and scored by the company selling the model. And for an agent that clicks through real software, even a small error rate means wrong entries in real ledgers. Cheap automation multiplies mistakes as efficiently as it multiplies output.

OpenAI also takes direct aim at Anthropic, saying Terra at max effort is over 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-selected benchmarks, so we will wait for independent testing before calling the price war. But the direction is clear: capable agents are getting commoditized fast.

The cybersecurity line is where we want to slow down. A model that meets OpenAI's own Critical cyber threshold is now the cheap, fast option, available broadly and built for operating computers. OpenAI says it keeps the same safeguards as its flagship and shows strong resistance to prompt injection in its tests. We take that seriously, but cheaper access to strong cyber capability is exactly the trend that worries defenders. If this scales the way the pricing suggests, attackers get a cost cut too, and the safeguards have to hold at much higher volume.

And the labor question is unavoidable. A model marketed for the communication and computer-based tasks that fill the workday is, by design, aimed at work people are paid to do now. We think the gains are real. We also think companies should be honest about whether the time saved goes to workers or simply to fewer of them.

I'm watching the system card, independent benchmark runs, and the first reports of Terra-based agents handling finance and back-office work in production.

Terra is OpenAI's efficiency-focused model line, sitting below GPT-6.1 Sol and GPT-6 Astra. OpenAI says Sol and Astra remain the better choice for the most complex coding tasks, while Terra more than doubles its predecessor on Terminal-Bench 4.0, solving 50.3 percent of tasks at max effort versus 21.5 percent.