OpenAI launches GPT-6.1 Terra for precise computer use
GPT-6.1 Terra is now available in ChatGPT Work, Codex, and the OpenAI API. It can use computers faster and costs less at standard API prices.
OpenAI has released GPT-6.1 Terra, a fast and efficient AI model that can understand images in detail, use computers accurately, and write in a clearer, more natural way. OpenAI says it is a major improvement over GPT-5.6 Terra. The company says it works especially well for writing proposals, working with technical drawings and forms, and editing videos.
On Agents' Last Exam, a test of how well AI agents use software to do work, OpenAI says Terra with medium reasoning scores 15 percentage points higher than GPT-5.6 Terra while costing about one-third as much. It also matches GPT-6 Astra while costing one-tenth as much per token. At the highest reasoning setting, OpenAI says Terra uses more than three times fewer tokens than Claude Opus 5.5 and costs one-fourth as much per token. This means the cost of completing a task is about nine times lower. OpenAI also says Terra completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.
BREAKING 🔥: OpenAI officially announced GPT-6.1 Terra!
— 🚨 AI News | TestingCatalog (@testingcatalog) October 26, 2026
> "At max effort, Terra is over three times as token-efficient as Claude Opus 5.5, and one-fourth its price per token, resulting in a ~9x lower cost per task."
> "Terra is also much quicker, completing tasks over three…
OpenAI says Terra scores better than GPT-5.6 Terra on MMMU-Pro at every reasoning setting. It also comes close to GPT-6 Astra and GPT-6.1 Sol while costing less. On OSWorld 2.1, a test of computer-use skills, Terra with medium reasoning scores higher than GPT-5.6 Terra's best score while costing much less for each task.
OpenAI tested how often Terra makes mistakes when given prompts designed to cause errors. With low reasoning effort, Terra's error rate was 16.7%, down from 53.6% for GPT-5.6 Terra. At the highest effort setting, its error rate fell to 4.3%. This is an 18-percentage-point drop and brings it close to GPT-6 Astra and GPT-6.1 Sol. OpenAI also says Terra completes 50.3% of Terminal-Bench 4.0 tasks at the highest effort setting, compared with the previously reported 21.5% for GPT-5.6 Terra. OpenAI says GPT-6.1 Sol and GPT-6 Astra are still better choices for the most difficult coding tasks.
GPT-6.1 Terra performs much better than GPT-5.6 Terra on Agents’ Last Exam, a test of how well AI agents use software at work.
— OpenAI (@OpenAI) October 26, 2026
It also matches GPT-6 Astra at one-tenth the price per token and finishes tasks more than three times faster on average than both Astra and GPT-6.1 Sol. pic.twitter.com/pzAmnvjBUi
OpenAI says Terra meets the Critical cybersecurity threshold in its Preparedness Framework, which it uses to assess serious AI risks. Terra also keeps the safety protections used for GPT-6 Astra. OpenAI says Terra follows instructions more closely than GPT-5.6 Terra, is more honest about its limits and obstacles, and resisted prompt injection in tests. Prompt injection is when someone tries to trick an AI into ignoring its instructions. OpenAI has published a system card with more details about the model.
GPT-6.1 Terra is now available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Its API model ID is gpt-6.1-terra. Standard API prices are $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. These prices are 50% lower than those for GPT-5.6 Terra and GPT-6.1 Sol. OpenAI says users can complete more tasks with the same usage allowance than with GPT-5.6 Terra, and many more than with GPT-6.1 Sol or GPT-6 Astra.
Sources and related information
- GPT-6.1 Terra system card addendum: OpenAI explains the model's safety tests and protections, including its classification at the Critical cybersecurity level.