OpenAI's GPT-6.1 Terra is built to do your busywork. Here's what it can do

The big story: OpenAI made a cheaper model for using your computer

On October 26, 2026, OpenAI launched GPT-6.1 Terra, which it describes as "a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing" (OpenAI, Oct 26, 2026).

Much of the work that fills a day happens on a screen: reading an invoice, typing it into accounting software, filling out a form, checking a drawing. OpenAI says Terra is built for that kind of task, and says its API prices make it practical to automate high-volume work like reading incoming invoices and entering the details into accounting software.

It is available now in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users (OpenAI; The Verge, Oct 26, 2026), and in the API under the model ID gpt-6.1-terra (OpenAI developer docs). It also became generally available the same day on Amazon Bedrock and Microsoft Foundry (both Oct 26, 2026).

Full explainer with pricing and limits: GPT-6.1 Terra, the cheaper AI built for computer use.

Why people are talking about it

The headline numbers come from OpenAI. They are OpenAI's own results, not independent tests:

• On Agents' Last Exam, a benchmark of professional computer tasks, OpenAI says Terra at medium effort scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost, and performs on par with GPT-6 Astra at one-tenth its price per token (OpenAI).
• OpenAI says Terra completes tasks over three times faster than both GPT-6 Astra and GPT-6.1 Sol (OpenAI).
• At max effort, OpenAI says Terra is over three times as token efficient as Claude Opus 5.5 at one-fourth its price per token, which it puts at a roughly 9x lower cost per task (OpenAI).
• On OpenAI's internal benchmark of error-inducing prompts, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra (OpenAI).

Standard API pricing is $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million (OpenAI developer docs). AWS says it supports a context window of up to 1 million input tokens and accepts text and image inputs (AWS, Oct 26, 2026).

One limit OpenAI states itself: for the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra "remain the better choice" (OpenAI). Terra is positioned as the everyday model, not the top specialist.

What this means for you

If you have been waiting for computer-use AI to get cheaper, this launch is worth a look. On OpenAI's numbers, a model that reads screens well and costs less per task makes it more practical to hand off repeatable work: data entry from PDFs, moving details between apps, first drafts of proposals.

A simple way to start:

1. Pick one repeat task you do on screen every week, like copying invoice details into a spreadsheet.
2. Run it once in ChatGPT Work or Codex with Terra, and watch each step.
3. Keep a human check on anything that sends, pays or deletes. Speed only helps if the result is right.

Yesterday we covered Anthropic's advice to fold most subagents into one agent that loads skills on demand: Anthropic says one AI agent beats a team of subagents. A lower-cost computer-use model is one more reason to try the single-agent setup on a routine task.

Also today: Anthropic files to go public

Anthropic filed a registration statement on Form S-1 with the SEC on October 26, 2026, and says it intends to list its Class A common stock on the Nasdaq Global Select Market under the symbol "ANT." The number of shares and the price range have not been determined (Anthropic, Oct 26, 2026; SEC EDGAR).

Figures from the filing, as reported by CNBC (CNBC, Oct 26, 2026):

• Revenue for the first half of 2026 was $16.29 billion, up 1,304% from $1.16 billion a year earlier.
• The net loss was $80.7 billion, with about $72 billion coming from accounting charges tied to changes in the value of convertible notes.
• Net revenue retention was around 900% as of September 30.
• More than 90% of contributions to Anthropic's codebase in September were written by Claude.

This is news, not investing advice. For the plain-English version, see Anthropic's S-1 explained.

New tools to try

• GPT-6.1 Terra (openai.com). OpenAI's new model for computer use, visual reasoning and writing, in ChatGPT Work, Codex and the API (OpenAI, Oct 26, 2026).
• Strawberry 1.0 (strawberrybrowser.com). An AI browser for non-technical teams, released as version 1.0 on October 26, 2026. Strawberry calls its assistants "companions." The Next Web reports that trained companions can now be shared across a team, and that companions pause for approval before sending emails, publishing posts or changing records. The browser is available for macOS and Windows, with a free plan and paid plans starting at $20 a month, according to The Next Web (Strawberry, Oct 26, 2026; The Next Web, Oct 26, 2026). Check Strawberry's own pricing page before you sign up.
• Riverside Agents (riverside.com/agents). Riverside launched four AI agents for video on October 26, 2026: a Producer, an Editor, a Clipper and a Marketer, rolling out on web and iOS. Riverside says nothing is published until you approve it, and that Agents are available on every Riverside plan, with a 14-day free trial of Riverside (Riverside, Oct 26, 2026; Riverside Agents).
• Unlimited OCR 2 (huggingface.co/baidu/Unlimited-OCR-2). Baidu's open-weight OCR model for very long documents. Its Hugging Face page lists an October 27, 2026 release and an MIT license, and describes multi-page parsing in a single request. The model card's examples run on an NVIDIA GPU, so it is aimed at developers (Hugging Face; MarkTechPost, Oct 26, 2026). Its benchmark scores are Baidu's own.

The bottom line

OpenAI says its new model can use a computer faster and for less than its earlier models. Those are vendor numbers, so start with one boring task, keep a human check on anything that matters, and see what it saves you.

This post has no affiliate links. See our affiliate disclosure and how we review.