OpenAI Launches GPT-6.1 Terra: Computer Use at Half the Price
3 min readOpenAI has launched GPT-6.1 Terra, a model built to operate software the way a person does: reading what is on screen, clicking through applications, filling in forms and finishing multistep tasks. The headline is not only capability but cost. Standard API pricing lands at $1 per million input tokens and $5 per million output tokens, half the price of both GPT-5.6 Terra and the larger GPT-6.1 Sol.
Where GPT-6.1 Terra fits in the lineup
OpenAI now runs a tiered family. GPT-6 Astra and GPT-6.1 Sol sit at the top for the hardest reasoning and coding work. The Terra line is the fast, efficient tier aimed at agents that have to run constantly and cheaply. GPT-5.6 Terra, the previous entry, was quick but error prone, which limited how far companies trusted it with unattended work.
What OpenAI is claiming
According to the announcement, summarized by TestingCatalog, GPT-6.1 Terra scores 15 points higher than GPT-5.6 Terra on Agents’ Last Exam, a test of how well agents use workplace software, at about a third of the cost per task. On Terminal-Bench 4.0 it reaches 50.3 percent at maximum effort, against a published 21.5 percent for its predecessor. On OSWorld 2.1 it beats GPT-5.6 Terra’s best score at a fraction of the cost.
Reliability is the other pillar. On OpenAI’s internal set of error-inducing prompts, the low-effort error rate dropped from 53.6 percent to 16.7 percent, and to 4.3 percent at maximum effort. OpenAI also says the model sticks to the scope it was given, discloses blockers more openly and resists prompt injection in testing. It completes tasks more than three times faster than Astra or Sol. In a pointed comparison, OpenAI says Terra is over three times as token efficient as Claude Opus 5.5 at one fourth the price per token.
The model is available now to ChatGPT Plus, Pro, Enterprise and Edu subscribers through ChatGPT Work and Codex, and to developers via the API as gpt-6.1-terra. Sam Altman wrote on X that the company is “especially excited for this new era of computer use.” OpenAI classifies Terra at the Critical cybersecurity threshold under its Preparedness Framework and ships a system card addendum alongside it.
Why it matters
Computer use agents have been the most expensive kind of AI to run, because every screen read and every click costs tokens. Cutting the per-token price in half while raising accuracy changes the arithmetic for back-office automation such as claims processing, procurement and data entry. It also raises the pressure on Anthropic and Google, whose agent products now face a cheaper rival that OpenAI is explicitly benchmarking against Claude.
The numbers are OpenAI’s own, so watch for independent runs on OSWorld and Terminal-Bench in the coming weeks. If they hold, Terra becomes the default engine for a lot of agent workloads.