OpenAI’s new model can work on your screen

Think about the dullest hour of your week. Copying bill details from a PDF into your accounts software. Filling the same vendor form again. Moving numbers from one screen to another.

This week, OpenAI released a model built for exactly that kind of work.

GPT-6.1 Terra is a new, faster model in ChatGPT. It can see what is on your screen and work across apps for you: clicking, typing and moving between them.

It works on your screen

OpenAI says it can fill in forms, log bills into accounting software, read technical drawings and draft proposals.

It makes fewer mistakes

On OpenAI’s own test of tricky questions, wrong answers fell from 53.6% to 16.7% at its lightest setting, compared with the older Terra model.

It costs half to build with

For companies building their own tools on it, OpenAI has cut the price to half that of the older Terra model.

For a growing business, this means the repetitive clicking can be handed over, while the checking and the decisions stay with you. A CA firm could use it to enter bills during the GST rush. An operations team could use it to fill vendor forms from emails.

These are OpenAI’s own numbers, so try it on one real task from your week before you trust it with more.

Who gets it: ChatGPT Plus, Pro, Enterprise and Edu users, from 26 October. Developers can use it through OpenAI’s API. OpenAI has not announced it for the Free or Go plans.

1Bengaluru’s Maya Research releases a free voice model for 12 languages

Maya Research has released Maya 2 Native, a voice model that speaks English and 11 Indian languages, including Assamese, Odia and Punjabi. It comes with 20 ready voices, and it can copy a voice from a clip of just 3 to 10 seconds and speak in another language.

It is free to use, including for business. Maya asks people to copy only voices they own or have permission to use.

Who gets it: developers and tech teams. You download it and run it on your own computer with a graphics card (12 GB or more is suggested).

2Gemini Notebook now works inside Gemini chat

Since 23 October, you can manage your notebooks without leaving the Gemini chat. Type @Gemini Notebook to make a new notebook, add files and links, or ask questions that are answered from your own sources.

It is handy for any work with many documents: vendor contracts, research for a pitch, or notes from a long project.

Who gets it: personal Google accounts and Google Workspace users. It is switched on by default in India.

3OpenAI asks scientists to check its AI’s biggest results

OpenAI says its AI models have produced new results in maths, computer science and physics. One is a claimed proof for the Yang-Mills problem, one of the seven Millennium Prize problems in mathematics.

Instead of declaring them settled, OpenAI has handed the work to independent groups of researchers to check and explain. There are no NDAs, its models take no academic credit, and it is giving $5 million to the Simons Foundation for workshops and conferences.

The claims still need expert checking. That is how it should work. AI does the heavy lifting, and people decide what holds up.

Who should care: anyone curious about how AI and human experts will work together on hard problems.

Until next Wednesday,

P.S. Know someone who loses hours to forms and spreadsheets? Forward them this note.

Run a one-person sales desk with Claude and Airtable. It takes about 20 minutes to set up.

Connect Claude to Airtable (an online sheet that works like a simple database), and it can act as your own small sales and marketing team. Here is a simple setup to try.

AI Workflow Accelerator: Cohort 3 is open

Registrations are open for Cohort 3 of our AI Workflow Accelerator for Professionals. Over four weeks of live sessions, you build practical AI workflows around the work you actually do, and leave with one you can use right away.

5,000+ professionals

Founders, business leaders and practitioners working through the AI shift together, one conversation at a time.