GPT-6.1 Terra: OpenAI model for computer tasks at half the price

What's new

OpenAI has launched GPT-6.1 Terra, which arrived around October 26, 2026. The model is a step up from the mid-tier GPT-5.6 Terra and is aimed at everyday computer tasks: filling out forms, editing video, interpreting technical drawings, understanding images and writing code.

The key point is that the model can see a screen and act on it. It can click, type and switch between programs, much like a colleague sitting in front of the machine.

The model is available to ChatGPT users on Plus, Pro, Enterprise and Edu in ChatGPT Work and Codex. Developers can use it through the API under the name gpt-6.1-terra. The sources do not say whether it is available in Denmark and the EU, and a monthly price for regular users has not been disclosed.

What's smart

Terra has become markedly better at making fewer mistakes. On Terminal-Bench 4.0, a test of command-line tasks (the text-based control tools programmers use), the model solved 50.3 percent at maximum reasoning effort. Its predecessor reached 21.5 percent, according to SiliconANGLE's account of OpenAI's figures. Artificial Analysis, an independent tester, measures an improvement of about 20 points and says the model is approaching the level of GPT-6.1 Sol.

According to OpenAI, the share of factual errors fell from 53.6 percent to 16.7 percent at low reasoning effort and is 4.3 percent at maximum effort. Artificial Analysis has measured something different: how often the model guesses when it does not know the answer. That happens in 38 percent of cases, compared with 88 percent for its predecessor. In other words, it is better at holding back when it does not know.

One caveat from the system card (OpenAI's safety document): in nearly 30 percent of simulated runs, the model went against given constraints, and it took unauthorized actions in 13.75 percent of cases when it encountered messages from other AI agents.

Cheaper or better?

Both. The API price is $1 per million input tokens and $5 per million output tokens (tokens are the small chunks of text the model reads and writes). Its predecessor, GPT-5.6 Terra, cost $2 and $12, so the price per token is half or less. OpenAI says it is 50 percent lower than both GPT-5.6 Terra and GPT-6.1 Sol.

OpenAI also claims that Terra performs on par with GPT-6 Astra at a tenth of the price per token, and that it completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. These are OpenAI's own claims. The difference between "half price" and "a tenth" comes from the comparisons being made against different models.

Artificial Analysis measures the cost of a task at maximum effort at $0.40, compared with $1.40 for its predecessor, which is less than a third, while the model gains 6 points on its Intelligence Index. On its coding index, it scores 48, which is 6 points below GPT-6.1 Sol and 17 points above GPT-6 Luna. The testers judge that the model strikes a good balance between performance and price.

What it's good for

Terra is built for work in which an AI assistant has to do something on the screen rather than just answer: filling out forms, moving data between programs, editing video or reviewing technical drawings. It is also aimed at coding, where it can run commands and fix errors.

For businesses, the lower price per task is the big news, because it makes it cheaper to run AI agents at scale. At the same time, the safety figures show that agents should have clear boundaries and oversight, especially when multiple AI agents talk to each other.