GPT-6.1 Terra lists a one-dollar input price

OpenAI’s API changelog dates GPT-6.1 Terra to October 26, 2026, and lists it on the Responses and Chat Completions APIs. The company announcement calls it a model for visual understanding and computer use, with writing the company describes as clearer, and says standard API prices are 50 percent lower than those of GPT-5.6 Terra and GPT-6.1 Sol. The posted rates are $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. The announcement says that starting that day, Plus, Pro, Enterprise, and Edu accounts can use it in ChatGPT Work and Codex. The API model name is gpt-6.1-terra.

The computer-use figures are the company’s own measurements. The page does not attach a table from an outside lab. On Agents’ Last Exam, at medium reasoning effort, the announcement says Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost, and matches GPT-6 Astra at one-tenth of Astra’s price per token. At maximum effort, the company says Terra is more than three times as token-efficient as Claude Opus 5.5, at one-fourth the price per token, for about a ninefold lower cost per task, and that it finishes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol.

The vision and desktop comparisons are also the company’s. On MMMU-Pro, a visual-reasoning test, the announcement says Terra beats GPT-5.6 Terra at every reasoning setting and comes close to GPT-6 Astra and GPT-6.1 Sol, which the company calls its strongest models, at a lower price. OSWorld 2.1 measures multistep work inside desktop applications. At medium effort, the company says Terra surpasses GPT-5.6 Terra’s best score at a fraction of the cost per task. The page does not say what that fraction is.

The factuality test is an internal set of prompts built to induce errors, not a public leaderboard. At low reasoning effort, the error rate fell from 53.6 percent with GPT-5.6 Terra to 16.7 percent with GPT-6.1 Terra. At maximum effort it fell 18 percentage points, to 4.3 percent, which the company says is close to Astra and GPT-6.1 Sol. On Terminal-Bench 4.0, maximum effort solved 50.3 percent of tasks, above the 21.5 percent the company cites as GPT-5.6 Terra’s published score. The announcement also says the most complex coding tasks are still a better fit for GPT-6.1 Sol and GPT-6 Astra. On Terminal-Bench-Science 0.1, the company says Terra at maximum reasoning surpasses GPT-6.1 Sol at lower cost and comes within five points of Opus 5.5 at less than one-fifth the estimated cost.

The announcement also names tasks inside those tests: turning architectural drawings into a three-dimensional model, reviewing an animated film shot by shot to make a time-coded log, fixing a webpage whose layout shifts while it loads, and tracking cell divisions in a time-lapse video. The same post mentions drafting proposals, reading invoices into accounting software, and editing video. Those are examples the company placed next to the scores. They are not problem statements from an outside organizer, and they do not show how the scores were graded.

Cybersecurity appears only as a result label. The company says Terra meets the Critical threshold in its preparedness framework and keeps the protections developed for GPT-6 Astra. The page does not describe how that test was run. The announcement also says the same ChatGPT usage allowance covers more tasks than with GPT-5.6 Terra, and substantially more than with GPT-6.1 Sol or GPT-6 Astra, without giving a multiple. Those scores, and the claim of more tasks per allowance, remain the company’s own accounting rather than an independent retest. The changelog line itself contains no scores.