A cheaper model for clicking things
OpenAI released GPT-6.1 Terra on Monday, a model it describes as combining visual reasoning with computer use — reading a technical drawing, filling in a form, editing a video. Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. OpenAI says that is 50% below the prices of both GPT-5.6 Terra, the model it succeeds, and GPT-6.1 Sol.
It is available from launch to Plus, Pro, Enterprise and Edu users inside ChatGPT Work and Codex, and through the API as gpt-6.1-terra. OpenAI says the same plan allowance covers more tasks on Terra than on GPT-5.6 Terra, and substantially more than on GPT-6.1 Sol or GPT-6 Astra.
The figures are OpenAI’s own
Every benchmark result in the announcement is vendor-reported. On Agents’ Last Exam, a set of professional tasks that includes turning architectural drawings into a 3D model, OpenAI says Terra at medium reasoning effort scores 15 percentage points above GPT-5.6 Terra at about a third of the cost, and performs on par with GPT-6 Astra at a tenth of its price per token. At maximum effort the company says Terra is more than three times as token-efficient as Anthropic’s Claude Opus 5.5 and a quarter of its price per token, which it works out to roughly a ninefold lower cost per task.
On Terminal-Bench 4.0, OpenAI reports that Terra at maximum reasoning effort solves 50.3% of tasks, against GPT-5.6 Terra’s published 21.5%. On Terminal-Bench-Science 0.1 it says Terra passes GPT-6.1 Sol at lower cost and comes within five points of Opus 5.5 at less than a fifth of the estimated cost. It also claims gains on MMMU-Pro and OSWorld 2.1 over GPT-5.6 Terra, without publishing the underlying scores in the post.
The most concrete claim is about errors. On an internal benchmark of error-inducing prompts, OpenAI says the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with Terra, and at maximum effort fell 18 percentage points to 4.3%. That benchmark is OpenAI’s own and is not published.
An independent index puts it 25th of 221
Artificial Analysis, which runs its own evaluations, scores GPT-6.1 Terra at maximum effort at 48 on its Intelligence Index, against a class median of 28. That places the model 25th of 221 on intelligence and 32nd of 120 on cost, with a blended rate of $0.73 per million tokens and $0.40 per Intelligence Index task.
The same page records one cost the launch post does not: Terra used 130 million output tokens to work through the index, against a median of 81 million. A model priced at a fifth of a rival can still spend more to get to an answer, and that is the number buyers of high-volume automation will want reproduced.
Critical in cybersecurity, at a tenth of the price
OpenAI states that GPT-6.1 Terra “meets the Critical threshold in cybersecurity” under its Preparedness Framework and “maintains the protections developed for GPT-6 Astra.” That is the notable line in the post. The safeguards built for the frontier model now ship with the cheap, fast one that anybody on a Plus plan can point at a browser.
OpenAI also says Terra stays within scope and discloses blockers more readily than GPT-5.6 Terra, and that it resisted prompt-injection attempts designed to divert it from a user’s task, pointing to the system card for detail. Those are evaluations the company ran on itself.
What to watch is whether independent reproductions of Terminal-Bench 4.0 and OSWorld 2.1 land near OpenAI’s numbers, and how the Critical classification interacts with the access rules OpenAI set out earlier this month for gateways reselling its models.
