OpenAI has introduced GPT-6.1 Terra, a model designed for AI agents that operate inside applications rather than primarily handling short conversational requests. The company says its standard API rates are 50% lower than GPT-6.1 Sol’s, positioning Terra for high-volume automation workloads where execution time and token costs are central concerns.
Terra can interpret images and technical drawings, fill in forms, prepare proposals and analyze video frame by frame. OpenAI positions it below GPT-6 Astra for the most demanding projects, but as a faster, less expensive option for repeated agent tasks.
Benchmarks and reliability claims
In OpenAI’s Agents’ Last Exam benchmark, which includes tasks such as turning architectural plans into 3D models and reviewing a film to create a time-stamped log, GPT-6.1 Terra reportedly scored 15 percentage points above GPT-5.6 Terra at medium reasoning effort while costing roughly one-third as much per task.
At maximum effort, OpenAI says Terra completes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol. Those results are company-reported benchmark figures rather than a guarantee for every automated workflow.
For coding, Terra achieved 50.3% on Terminal-Bench 4.0, compared with 21.5% for GPT-5.6 Terra. OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra for complex software projects where capability takes priority over speed and cost.
The company also reports lower error rates on an internal set of questions specifically designed to induce incorrect answers. At low reasoning effort, the rate fell from 53.6% to 16.7%, while maximum effort produced a 4.3% rate. Because the evaluation targets failure cases, those figures should not be treated as a general error rate for all chats, documents or code.
OpenAI says Terra follows supplied limits more reliably and is more resistant to malicious instructions embedded in analyzed content. That is particularly relevant for agents processing emails, documents and web pages, where prompt-injection attempts can be present.
Availability and API pricing
GPT-6.1 Terra is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscriptions. API access is listed under the gpt-6.1-terra model identifier.
- Standard input: US$1 per million tokens.
- Reused cached input: US$0.05 per million tokens.
- Output: US$5 per million tokens.
- Batch and Flex pricing: half the standard rates.
- Fast mode: double the standard rates.
Terra’s placement between the flagship GPT-6 Astra and cheaper models makes it a targeted option for developers running large numbers of agent tasks. Its practical standing against Sol and Astra will depend on whether the company’s speed, reliability and cost claims hold up across production workloads.
