OpenAI releases GPT-6.1 Terra with lower prices for visual AI workflows

OpenAI released GPT-6.1 Terra on October 26, positioning the model as a less expensive option for work that combines visual understanding, writing and actions in software. Its launch announcement says access is starting in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, as well as through the API.

This is more than an announcement of future research access: the API changelog records the October 26 release for both the Responses and Chat Completions endpoints. Developers can request gpt-6.1-terra; applications using its tool-calling features need the Responses API.

A lower-priced model for visual workflows

The model documentation lists standard prices of $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. Those are token rates, not a fixed price per completed task. Long exchanges, repeated attempts and tool use can change the bill.

The documented context window is 1,050,000 tokens, with a maximum output of 128,000 tokens. Terra accepts text and images and produces text; the model page does not list audio or video support. Requests exceeding 272,000 input tokens receive higher long-context pricing, so the headline rates should not be treated as universal.

OpenAI's model comparison lists GPT-6.1 Sol at $2 per million input tokens and $10 per million output tokens. Terra's standard rates are half those figures. The practical buying question is whether a cheaper model completes the particular workflow reliably, not simply whether its first response costs less.

What the performance claims do—and do not—show

OpenAI says Terra improves on GPT-5.6 Terra in visual reasoning, computer use and writing. On the company's Agents Last Exam evaluation, it reports a roughly 15-percentage-point improvement at medium reasoning effort at about one-third of the cost. These are vendor-reported evaluation results, not independent measurements by PromptsZone.

The release also illustrates a distinction that matters for agent software: reading a screenshot correctly is not the same as safely completing a multi-step action. A workflow can still fail through a mistaken click, an incorrect assumption about the user's goal or an unsuccessful recovery from an earlier error. Lower prices make evaluation cheaper; they do not remove the need for it.

Availability is only the start of deployment

The October 26 safety-card addendum classifies Terra as Critical for cybersecurity capability and High for biological and chemical capability under OpenAI's framework. Those are the developer's capability classifications, not a statement that every ordinary use is dangerous. The document describes safeguards and warns that research-environment evaluations can differ from production behavior because of prompting, tools and reasoning settings.

For teams considering a switch, a sensible next step is a small comparison on their own representative tasks, including unsuccessful runs—not an immediate replacement based on a leaderboard. OpenAI's model-selection guidance likewise emphasizes task-specific evaluation. Review checkpoints, limited permissions and clear stopping conditions remain important when an application can act on a user's behalf.

Terra's significance is the combination of a new visual-agent option and lower standard token prices. Whether that becomes a lower cost per useful result will depend on the application, its error tolerance and the amount of human review required.

Photo: Sam Altman at The Grove, November 17, 2022—archival, not this launch. Village Global, via Wikimedia Commons, CC BY 2.0; resized.