OpenAI rolls out GPT-6.1 Terra to ChatGPT Work, Codex and API at lower token prices

OpenAI releases GPT-6.1 Terra across ChatGPT, Codex and the API

OpenAI has made its new GPT-6.1 Terra model broadly available, shipping it simultaneously to ChatGPT Work, the Codex environment and the public API. According to dev.to, the model is open to Plus, Pro, Enterprise and Edu subscribers in ChatGPT Work and Codex, while developers can call it directly using the identifier gpt-6.1-terra.

The model targets work that combines visual information with actions and written output. OpenAI's announcement, as relayed by dev.to, names reading technical drawings, preparing proposal drafts and video editing as representative examples, and the API documentation adds catalog matching and form interpretation to the list.

A cheaper option, not a universal replacement

The positioning is the notable part. Rather than pitching Terra as a successor for every workflow, OpenAI presents it as a lower-cost choice specifically for jobs that need visual understanding and precise computer use.

The company claims GPT-6.1 Terra beats its predecessor GPT-5.6 Terra on key visual-reasoning tasks and comes close to GPT-6 Astra on the same work while costing far less. These are OpenAI's own performance and cost figures, and dev.to notes that teams should validate them against their own inputs, tools and quality requirements rather than taking them at face value.

Pricing details

API access is priced at $1 per million input tokens, $5 per million output tokens and $0.05 per million cached input tokens, according to the announcement. The cached rate stands out for workflows that repeatedly send the same context, such as reference documents, catalog data or standard form layouts, since it charges one-twentieth of the standard input price for that recurring material.

Three routes in

Availability spans three surfaces, which shapes how teams can try the model. ChatGPT Work suits hands-on, app-based evaluation; Codex serves development work; and the API allows integration into existing applications and automated pipelines. The right route depends on whether the goal is individual productivity, coding assistance or a repeatable operational process.

Migration and validation

For teams already running GPT-5.6 Terra, dev.to cautions against treating the release as a pure pricing update: output style, how the model interprets visual inputs, and how it interacts with tools can all shift and affect downstream systems. The suggested approach is to replay representative inputs from an existing workflow and compare results against the same acceptance criteria, keeping human approval in place for anything customer-facing or irreversible.

A related point from the report: lower token prices do not automatically translate into lower operating costs. The metric that matters is cost per completed, acceptable task. A cheaper model that needs retries and manual correction can end up costing more than a pricier one that finishes work reliably; conversely, a low-cost model can improve the economics of high-volume processes where human review stays manageable.

OpenAI also says it has improved safety and alignment in this release and published a System Card describing deployment safeguards. That documentation is relevant for implementations handling customer, financial or operational data, though it does not replace defining, per workflow, what the model may do on its own, what requires review, and how errors get handled.

Why it matters

The release pairs unusually broad access with aggressive pricing on the visual-reasoning and computer-use front. If OpenAI's claims hold up, tasks that were previously too expensive to automate at volume, such as extracting data from forms, matching products across catalogs or producing first drafts of proposals, become realistic candidates for AI assistance.

The sensible response for developers is a bounded pilot: pick one repeatable task with known volume and a clear quality bar, then measure completion time, output acceptance rate, review effort and total cost per finished task. That turns OpenAI's cost-positioning claims into an evidence-based decision about whether GPT-6.1 Terra belongs in production.