GPT-6.1 Terra: OpenAI halves the price of its mid-tier model and targets computer use
Available since October 26 in ChatGPT, Codex and the API, GPT-6.1 Terra costs $1 per million input tokens. Its performance gains have, for now, been measured only by OpenAI.
On October 26, OpenAI launched GPT-6.1 Terra, the new model in its mid-tier line, sitting between its most powerful models (GPT-6 Astra, GPT-6.1 Sol) and high-volume use cases. It is available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers through the API as gpt-6.1-terra. Amazon made it available on Bedrock the same day.
What really changes: the price
The most solid fact in this announcement is the pricing. The API charges $1 per million input tokens, $0.05 for cached input and $5 for output. OpenAI says these standard prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol. According to AWS’s listing, the model accepts text and images, with a window of up to 1 million input tokens.
For high-volume workloads (reading invoices, entering data into business software, agents chaining many steps), the price per token is what drives the bill. OpenAI itself highlights this kind of office automation rather than cutting-edge research tasks.
Computer use and code: in-house numbers
OpenAI presents Terra as a computer-use model: understanding a screen, filling in forms, interpreting technical drawings. On Agents’ Last Exam, a benchmark of professional tasks, the company reports 15 points more than GPT-5.6 Terra at medium reasoning effort, for about a third of the cost, and a level comparable to GPT-6 Astra at a tenth of its price per token. It also claims Terra completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.
In coding, OpenAI claims 50.3% of tasks solved on Terminal-Bench 4.0 at maximum effort, versus 21.5% for GPT-5.6 Terra. The company itself acknowledges that for the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice.
On reliability, OpenAI cites an internal benchmark of prompts designed to trigger errors: the error rate reportedly falls from 53.6% to 16.7% at low effort, and to 4.3% at maximum effort. Since this test is not public, it cannot be reproduced.
The limits
All performance scores come from OpenAI, with its own settings. No independent measurement was available at the time of publication. The cost comparisons with Claude Opus 5.5 (“about 9 times cheaper per task” at maximum effort) depend on the number of tokens consumed, and therefore on the chosen task. Finally, OpenAI says the model meets the “Critical” cybersecurity threshold of its Preparedness Framework and keeps the protections developed for GPT-6 Astra: a capability level that comes with safeguards likely to refuse some cybersecurity requests.
Our take
The price cut is concrete and verifiable on the price list; the performance gains still need third-party confirmation. For high-volume office automation, Terra deserves an immediate test against its predecessor. For complex code, OpenAI itself points to its more expensive models: Terra is not pitched as the best, but as the most cost-effective.
Sources
- OpenAI — Introducing GPT-6.1 Terra Primary source
- AWS — OpenAI GPT-6.1 Terra is now generally available on Amazon Bedrock Primary source — availability date and context window
A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.
