Most of the work that fills an office day is not frontier mathematics. It is reading an invoice and typing its numbers into accounting software, filling out a form, checking a drawing, or trimming a video. On October 26, 2026, OpenAI released a model aimed squarely at that kind of work: GPT-6.1 Terra, a fast, mid-tier model that combines visual reasoning with the ability to operate a computer, at a standard API price 50% lower than the model it replaces.
The launch says a lot about where the AI market is heading. The race is no longer only about the single smartest model; it is increasingly about how much useful work a model can do per dollar and per second.
What GPT-6.1 Terra is
In its announcement, OpenAI describes Terra as “a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing.” It succeeds GPT-5.6 Terra and sits below the company’s heavier GPT-6.1 Sol and GPT-6 Astra models. OpenAI’s examples are deliberately ordinary: drafting proposals, interpreting technical drawings, filling out forms and editing videos.
The model went live the same day for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and for developers in the API as gpt-6.1-terra. According to a weekly roundup by The Model Press, OpenAI’s model page lists a context window of about 1.05 million tokens with text and image input, and AWS made Terra generally available on Amazon Bedrock on launch day.
The numbers OpenAI is highlighting
Three sets of figures stand out in OpenAI’s post. All of them are the company’s own measurements, so independent testing will matter.
- Computer use: On a benchmark of professional tasks called Agents’ Last Exam, OpenAI says Terra at medium reasoning effort scores 15 percentage points higher than GPT-5.6 Terra at about one-third of the cost, and performs on par with GPT-6 Astra at one-tenth of Astra’s price per token. It also claims Terra completes tasks more than three times faster than Astra and GPT-6.1 Sol.
- Fewer factual errors: On an internal set of error-inducing prompts, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7%, and at maximum effort it fell to 4.3%.
- Coding: On Terminal-Bench 4.0, which tests coding agents working from the command line, Terra solves 50.3% of tasks at maximum effort, versus 21.5% for its predecessor. OpenAI still recommends Sol and Astra for the hardest coding jobs.
Pricing is the headline for developers: OpenAI lists standard API rates of $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens.
Why “computer use” is the real story
“Computer use” means a model looks at a screen, understands what is on it, and then clicks, types and navigates the way a person would. That matters because so much business software has no clean API. If an AI can operate the same interfaces people already use, automation stops depending on custom integrations.
OpenAI’s own framing makes the economics clear: lower prices make it “practical for developers to automate high-volume work, like reading incoming invoices and entering their details into accounting software.” A model that is slightly less brilliant but far cheaper and faster can be the better choice when a task repeats thousands of times a day. Decrypt summed up the trade-off bluntly, calling Terra a mid-tier model that now operates apps and screens more like a person would, at half the price.
Safety and reliability
Giving software the ability to act raises the stakes on mistakes. OpenAI says Terra keeps the safeguards developed for GPT-6 Astra, is more likely than its predecessor to stay within scope and to disclose limitations or blockers to users, and showed strong resistance in its prompt-injection tests, where hidden instructions try to divert an agent from its task. More detail is in the company’s system card.
Those claims are encouraging, but any team deploying an agent that clicks buttons on its behalf should still keep a human review step for actions that move money, send messages or change records.
Practical takeaways
- Match the model to the job. Use Terra for high-volume, well-defined tasks such as data entry, form filling and document review; keep heavier models for complex reasoning and difficult code.
- Benchmark on your own workflow. Vendor benchmarks are a starting point. Run a pilot on a few hundred real examples and measure accuracy, speed and cost per completed task.
- Design for oversight. Log every action an agent takes and require confirmation for irreversible steps.
- Watch total cost, not token price. A cheaper model that needs fewer retries and finishes faster can lower the true cost per task even more than the price sheet suggests.
The bottom line
GPT-6.1 Terra is not OpenAI’s most powerful model, and it is not meant to be. It is a bet that the next wave of AI value will come from dependable, affordable automation of everyday digital chores. If its real-world performance matches OpenAI’s numbers, the biggest change may be invisible: thousands of small, repetitive tasks quietly moving from people’s to-do lists to software.