GPT-6.1 Terra brings cheaper computer work
OpenAI’s new model targets visual tasks and everyday workflows. Test the finished result, including correction time, before handing it more responsibility.
OpenAI released GPT-6.1 Terra on October 26, aiming to make visual reasoning and computer work faster and less expensive. For people using ChatGPT Work or Codex, the useful question is whether it can finish a familiar task with less waiting and less correction.
OpenAI says Terra is available to Plus, Pro, Enterprise and Edu users in those products, and that subscription allowances cover more work than with its larger models. The announcement does not promise a fixed number of tasks: what you ask the model to do still matters.
A cheaper option to test
The API model page lists standard prices of $1 per million input tokens and $5 per million output tokens, with cached input at $0.05. Those are token rates, not the complete cost of an automated workflow. Tool charges, retries and human checking can change the total.
OpenAI describes uses such as matching product photographs to catalog entries and working with application screens. It reports stronger computer-use results than GPT-5.6 Terra, while continuing to recommend Sol or Astra for its hardest coding tasks. These are the developer’s evaluations; we have not independently benchmarked Terra.
Simon Willison’s launch-day assessment identifies the same practical niche: repetitive computer work can benefit from a less expensive model. His post includes a small drawing experiment, not a comprehensive reliability test.
Start with a task you can check
Our suggestion is to compare Terra with your current model on a small set of completed, non-sensitive tasks. For a catalog workflow, include visually similar products and examples with missing information. Keep the original inputs and known answers.
Record whether the result is correct, how long it takes, and how much correction it needs. A faster first response is useful only if the finished work also improves. For developers, the current GPT-6 guide recommends comparing representative images and workflows at the same reasoning effort. Terra’s tool calls require the Responses API.
Capability still needs boundaries
OpenAI’s system-card addendum classifies Terra as Critical in cybersecurity and High in biological and chemical capability under its framework. Those are capability classifications, not independent assurances of safety. The card also cautions that its evaluation environment can differ from production ChatGPT.
Use a bounded task, restrict connected accounts to what it needs, and review consequential changes before they leave your workspace. Keep a working fallback. Lower prices make experimentation easier; they do not remove the need to inspect what the agent actually did.
Our model-selection guide and Everyday use resources provide further reading.