OpenAI releases GPT-6.1 Terra: $1/$5 per million tokens and a focus on computer use

OpenAI has released GPT-6.1 Terra, a fast, inexpensive model that the company pitches as a tool for the working day. It costs $1 per million input tokens and $5 per million output tokens. That is nearly half the price of GPT-5.6 Terra and GPT-6.1 Sol, and roughly ten times more than the entry-level GPT-6 Luna. The announcement emphasizes detailed visual reasoning, precise computer control and more natural writing.

What OpenAI put in the release headline

Simon Willison, who writes up every model release, noted what the announcement stresses. This time it is not math or coding benchmarks. OpenAI describes Terra as a model for the tasks that fill a working day: drafting business proposals, interpreting technical drawings, filling in forms and editing video. In the company's words, it is a significant step forward from GPT-5.6 Terra.

The logic is practical. Computer use, meaning the model works through interfaces instead of an API, usually turns into long, repetitive scripts: open a page, find a field, enter data, check the result. Such tasks handle a high per-token price and high latency poorly. A fast mid-tier model suits them better than a flagship.

Four models, and the question of which to pick

OpenAI's current frontier lineup looks like this:

• GPT-6 Astra: the top model in the lineup. According to Ted Sanders of OpenAI, Terra beats it at technical communication and holds even on vision and computer use, but falls behind on most other tasks.
• GPT-6.1 Sol: a tier above Terra. It costs nearly twice as much, but is stronger overall.
• GPT-6.1 Terra: the middle of the lineup, at $1 per million input tokens and $5 per million output tokens. Strengths: vision, computer control, clear writing, speed.
• GPT-6 Luna: the cheapest model. Roughly ten times cheaper than Terra, but noticeably weaker at vision, computer use and communication.
• GPT-5.6 Terra: the previous generation of Terra, which the new version undercuts on price by nearly half.

Willison compares Terra's position to Sonnet in Anthropic's family. It is neither the cheapest model nor the most expensive, so choosing it takes more calculation. You pick the flagship when you need maximum quality. You pick the smallest model when the token bill matters. The middle model turns out to be the right choice only for specific types of tasks.

Advice from an OpenAI engineer

Ted Sanders answered a question about positioning in a Hacker News comment. In his words, Terra is probably better than Astra and Sol at technical communication, on par on vision and computer use but noticeably faster, and slightly worse at most other tasks.

For developers using the API, he calls it a convenient middle option: cheaper than Sol, but much smarter than Luna, especially at vision, computer use and communication. For ChatGPT subscribers, by his own account, the release is not a bombshell. If you keep hitting usage limits, Terra can help you last a bit longer. If you need clear writing or faster computer work, it is worth selecting manually.

Such candor is rare in product announcements. OpenAI is effectively admitting that Terra does not replace the senior models but fills a specific niche.

A pelican for 11 cents

Willison ran Terra through his test of drawing an SVG pelican on a bicycle at every reasoning level. At the heaviest level, max, the generation cost 11.3161 cents and took 4 minutes 11 seconds. The prompt contained 16 input tokens, and the response ran to 22,629 output tokens. He rated the result positively: the bicycle frame has the right shape, the legs and pedals sit on both sides of it, and the color scheme matches other GPT-6 and GPT-6.1 models.

The example shows how cost builds up in practice. There are almost no input tokens, and nearly all of the price goes to reasoning and output. For models with reasoning modes, output tokens at $5 per million determine the final bill.

Willison also put together a comparison grid of pelicans for every reasoning level of Astra, Sol, Terra and Luna.

Where Terra pays off

The most obvious scenario for Terra is automating office routine through interfaces: filling in forms, processing documents with drawings and diagrams, preparing draft proposals, editing video. These are all tasks where speed and stability matter, while depth of reasoning is not always needed.

Developers should run their own computer-use scenarios on Terra alongside Sol and compare the cost per successfully completed task. If quality is close and speed is higher, switching is justified. For tasks that demand maximum general intelligence, it is better to stay with Astra or Sol. Luna fits bulk classification and simple queries. Subscribers should switch to Terra where clarity of text matters, or when their limits on the senior models run out.