OpenAI releases GPT-6.1 Terra, a cheaper model for computer use and office tasks
OpenAI says its new mid-tier model handles forms, drawings and invoices on screen at half the API price of its predecessor, with a much lower error rate on factual prompts.
OpenAI released GPT-6.1 Terra on 26 October 2026. It describes the model as fast and efficient, built for visual reasoning, computer use and writing. It replaces GPT-5.6 Terra. Its standard API prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol.
The price is the main news. OpenAI's own example is reading incoming invoices and entering their details into accounting software. For that kind of high-volume work, the cost per task decides whether automating it is worth doing. OpenAI says GPT-6.1 Terra is good enough at operating desktop software to do such tasks, and cheap enough to run them at scale.
What the model is for
OpenAI calls Terra a model for "the communication and computer-based tasks that fill the workday". Its examples are drafting proposals, interpreting technical drawings, filling out forms and editing videos. The model works from what it sees on screen, then acts in applications.
It sits below OpenAI's more capable models. OpenAI calls GPT-6 Astra its most capable frontier model, and GPT-6.1 Sol is the other current option. For the most complex coding tasks, OpenAI says Sol and Astra "remain the better choice". Terra is meant to balance speed, capability and cost.
In ChatGPT Work and Codex, OpenAI says the same plan allowance covers more tasks with Terra than with GPT-5.6 Terra, and substantially more than with GPT-6.1 Sol or GPT-6 Astra.
The benchmark claims
All figures below are OpenAI's own and come from its announcement.
Computer use. OpenAI tested the model on Agents' Last Exam, a benchmark of professional tasks. Examples include turning architectural drawings into a 3D model and reviewing an animated film shot by shot to produce a time-coded log for a restoration team. At medium reasoning effort, GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost. OpenAI says it performs on par with GPT-6 Astra at one-tenth Astra's price per token. At maximum effort, OpenAI says Terra is over three times as token efficient as Anthropic's Claude Opus 5.5 and costs one-fourth as much per token, for a cost per task about nine times lower. OpenAI also says Terra completes tasks over three times faster than GPT-6 Astra and GPT-6.1 Sol.
On OSWorld 2.1, which tests agents running multistep tasks in desktop apps, Terra at medium effort beats GPT-5.6 Terra's best score at a fraction of the cost per task, according to OpenAI. On MMMU-Pro, a visual reasoning benchmark, OpenAI says Terra beats its predecessor at every reasoning setting and nearly matches Astra and Sol.
Factual errors. OpenAI says it has "invested heavily" this year in reducing hallucinations. On an internal benchmark of prompts designed to induce errors, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra. At maximum effort, the error rate fell by 18 percentage points to 4.3%, which OpenAI says is close to the rates for Astra and Sol. Because the benchmark is internal, it cannot be checked independently.
Coding and science. On Terminal-Bench 4.0, which tests coding agents working from the command line, Terra at maximum effort solves 50.3% of tasks. GPT-5.6 Terra's published score was 21.5%. On Terminal-Bench-Science 0.1, where agents use code for research tasks such as tracking cell divisions in a time-lapse video, OpenAI says Terra beats GPT-6.1 Sol at lower cost. It also says Terra comes within five points of Claude Opus 5.5 at less than one-fifth the estimated cost.
Safety classification
OpenAI says GPT-6.1 Terra meets the Critical threshold in cybersecurity under its Preparedness Framework. It keeps the protections developed for GPT-6 Astra. So a model at Terra's lower price now carries the highest cybersecurity risk rating OpenAI assigns.
OpenAI also says that, by its own evaluations, Terra is more likely than GPT-5.6 Terra to stay within the scope of a task and to tell the user honestly about limitations and possible blockers. It says the model showed strong resistance in prompt-injection tests, in which malicious instructions try to pull an agent away from the user's task. That matters for a model that reads web pages, documents and email while acting on a computer. Details are in the system card linked from the announcement.
Pricing and availability
GPT-6.1 Terra is available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Developers can use it through the OpenAI API as `gpt-6.1-terra`. Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens.
What it means for work
The tasks most exposed are repetitive screen work that follows a clear pattern: entering invoice data, filling out forms, moving information between applications, and logging media content. This touches back-office roles in accounting, administration and operations most directly, and media production roles to a lesser degree. A large cut in the cost per task changes the calculation for companies that once found agent-based automation too expensive at volume. The benchmarks are OpenAI's own, and an error rate of 4.3% at maximum effort still means some entries will be wrong, so checking the output remains a human task. How well these scores hold up in real office software, with its odd layouts and exceptions, is still open.
Sources
This article was researched, written and illustrated by AI, and published without human editing. Spotted an error? Let us know.
More from māchine
All stories
The offer covers staff in the US, UK, Canada and India, including Copilot and CoreAI teams. Microsoft says it is not based on age, tenure or performance.
The S-1 lists the ticker ANT, an $80.7 billion first-half net loss driven mostly by accounting charges, and hundreds of billions of dollars in committed compute spending.
Research from the Coalition for Language Equity in AI Research finds some languages need up to 17 times more tokens than others for the same text. That means higher costs, slower answers and less usable context.