GPT-6.1 Terra: OpenAI's cheaper model for computer use

On October 26, OpenAI released GPT-6.1 Terra, a new “mid-tier” model built for computer use. According to the announcement, the model “understands what’s on screen, takes action across apps, writes naturally”, and in the API it costs half as much as its predecessor, GPT-5.6 Terra.

What’s new in GPT-6.1 Terra?

OpenAI now has four model tiers: from the cheapest, Luna, through Terra and Sol, up to Astra. As 9to5Mac writes, last month OpenAI refreshed Sol and Luna but skipped Terra. Now the middle model has caught up with the rest of the family.

According to the model card, GPT-6.1 Terra:

1. Costs $1 per million input tokens and $5 per million output tokens, while cached tokens are $0.05.
2. Has a context window of 1,050,000 tokens and up to 128,000 output tokens.
3. Accepts text and images and responds with text.
4. Has six reasoning levels, from “none” to “max” (the default is “medium”).
5. Supports, among other things, computer use, MCP, web search, the code interpreter and hosted shell in the Responses API.
6. Has a knowledge cutoff of June 1, 2026.

The use cases OpenAI lists (according to 9to5Mac) are fairly ordinary office work: writing proposals, reading technical drawings, filling in forms and editing video. In other words, the model is meant not only to write but also to click through apps for us.

How does it do in independent tests?

Here we already have the first numbers from outside OpenAI. In the Artificial Analysis Intelligence Index, GPT-6.1 Terra (in max mode) scores 48 points, against a median of 28 for comparable models, which puts it in 25th place out of 221 ranked models.

The price-to-score ratio is more interesting. According to Artificial Analysis, one task from their index costs $0.40 with this model, and its token prices are half the median for its class. There is, however, the other end of the same measure: during the test the model generated 130 million tokens, against a median of 81 million, so it likes to write at length (and output tokens cost five times as much as input tokens).

Why is it making waves?

The launch post quickly reached the front page of Hacker News. The reason is simple: mid-tier models are the ones we use most often, in ChatGPT and in agents that run for hours on end. Every price cut paired with a higher score translates directly into how much work an agent can do for the same money.

The second thing is the focus on computer use. A model that understands the screen and works across many apps, and is cheap on top of that, is exactly what background agents need. Cheaper models from other companies have recently taken a similar direction, e.g. MiniMax M3.1 Flash.

How to try GPT-6.1 Terra?

According to OpenAI, the model is rolling out now:

1. In ChatGPT Work and Codex for users on the Plus, Pro, Enterprise and Edu plans. Usage limits there are higher than with GPT-5.6 Terra and much higher than with GPT-6.1 Sol or GPT-6 Astra.
2. In the API as gpt-6.1-terra.

For deep research and hard programming, OpenAI still recommends the stronger GPT-6.1 Sol or GPT-6 Astra. Terra is meant to be the model “for a full day of work”, not for the hardest tasks. (if you use OpenRouter with EU routing, check availability: according to OpenRouter the model is not yet available in the EU region)

It’s also worth remembering that OpenAI regularly retires older models, as I wrote about the retirement of GPT-4o from the API, so migrating to the newer Terra is coming sooner or later anyway.

FAQ

How much does GPT-6.1 Terra cost?

In the API, $1 per million input tokens, $0.05 for cached tokens and $5 per million output tokens. According to OpenAI, that’s 50% less than GPT-5.6 Terra.

Is GPT-6.1 Terra available in free ChatGPT?

According to OpenAI’s announcement, the model is coming to users on the Plus, Pro, Enterprise and Edu plans in ChatGPT Work and Codex. The announcement doesn’t mention the free plan.

How is Terra different from Sol and Astra?

Terra is the mid tier in the OpenAI family: faster and cheaper, with higher limits. Sol and Astra are stronger models that OpenAI recommends for hard programming and deep research.