GPT-6.1 Terra

GPT-6.1 Terra is OpenAI's October 26, 2026 model release for high-volume professional work involving visual reasoning, computer use, coding, and writing. Its API identifier is gpt-6.1-terra. OpenAI says the model is available in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, as well as in the API. It is a new sibling, not a rename of GPT-6.1 Sol or GPT-6 Astra.

What it is for

• Visual work that ends in an action: OpenAI positions Terra for reading visual material, then carrying out a computer task, such as extracting invoice details into accounting software.
• Cost-sensitive agent loops: The launch price is $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens.
• Coding and factual tasks: OpenAI reports improvements over GPT-5.6 Terra on its internal factuality evaluation and Terminal-Bench 4.0. Those are vendor evaluations, not a promise about an individual codebase.
• Computer use with guardrails: OpenAI says Terra meets its Critical cybersecurity preparedness threshold and inherits GPT-6-series alignment protections. This does not remove the need for application-level permissions and review.

Practical workflow

Start with a bounded job: give Terra a test document or staging browser, a narrow action list, and an approval step before consequential writes. Use a representative evaluation set that includes both clean inputs and the kinds of visual ambiguity your team sees. Record completion quality, retries, latency, token use, and the rate at which a human rejects an action. Compare the result with GPT-6.1 Sol and GPT-6 Luna on cost per completed task, not just price per token.

Pricing and access

OpenAI's launch announcement lists standard API pricing of $1 input, $0.05 cached input, and $5 output per 1M tokens. Availability and product allowances can change, so check the official announcement and the current API pricing page before committing a production budget.

Limitations and alternatives

The benchmark results and price comparisons in the announcement are OpenAI's own measurements. They do not establish reliability for every desktop, browser, file format, or tool. A computer-use model can still misread a screen, follow untrusted page text, or choose an incorrect action. Keep sensitive systems behind least-privilege credentials, explicit confirmation, and an audit trail.

• GPT-6.1 Sol: choose it when your evaluation needs a more capable complex-work tier.
• GPT-6 Astra: choose it for the hardest work where model cost is secondary.
• GPT-6 Luna: consider it for lower-cost, high-volume jobs that need less capability.

FAQ

Is Terra a general permission to automate a desktop?

No. It is a model capability. The surrounding application must still constrain credentials, tools, and approvals.

What should I test first?

Use a non-sensitive, reversible workflow and compare end-to-end task success, human interventions, and total cost with your current model.

Conclusion

GPT-6.1 Terra is a new OpenAI option for teams that need visual reasoning and computer use without using the highest-priced GPT-6 tier for every task. Begin with the official release, then validate it on a permission-bounded workflow before routing production traffic.