GPT-6.1 Terra: Cheaper AI Agents for Australian Business

GPT-6.1 Terra is OpenAI’s new fast, lower-cost model for computer use, visual reasoning and everyday writing, and it is available now in ChatGPT Work, Codex and the OpenAI API. For Australian businesses, the interest is practical: OpenAI is pitching it at the screen-based admin that fills a working day, such as reading incoming invoices, filling in forms and drafting proposals, at half the standard API price of the model it replaces. Here is what launched, where it fits, and what to check before you let an AI agent loose on your systems.

What OpenAI released

OpenAI introduced GPT-6.1 Terra on 26 October (US time, early on 27 October in eastern Australia), describing it as “a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing”. The Verge reported the same day that it is rolling out to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. The key facts from OpenAI’s announcement and its developer model page:

• Model ID: gpt-6.1-terra, in the Responses and Chat Completions APIs and the Batch API. Tool calling needs the Responses API.
• Price: US$1 per million input tokens and US$5 per million output tokens at standard rates, which OpenAI says is 50 per cent lower than GPT-5.6 Terra and GPT-6.1 Sol. Batch and Flex processing are half that again, and fast mode costs double.
• Inputs: text and images in, text out, with a context window of just over one million tokens. It does not take audio or video.
• Tools: computer use, file search, web search, code interpreter and remote MCP servers, among others.

OpenAI says Terra is far quicker than its larger GPT-6 Astra and GPT-6.1 Sol models and comes close to them on its own visual reasoning and computer-use tests, while recommending Sol or Astra for in-depth research and the most complex coding. These are vendor-reported results, so treat them as a reason to trial it, not a buying decision.

Where GPT-6.1 Terra fits in an Australian business

Terra is aimed at work that happens on a screen rather than in a chat window. OpenAI’s own examples include reading invoices and entering their details into accounting software, interpreting technical drawings, filling out forms and matching product photos to catalogue entries. Realistic first trials for a small or mid-sized business include:

• Accounts payable admin: reading supplier invoices and preparing entries for a person to approve, rather than posting them directly.
• Legacy systems without an API: moving information between an old desktop or web system and a newer one where no integration exists.
• Catalogue and stock work: checking product photos and descriptions against your catalogue before they go live online.
• Drafting: first drafts of quotes, proposals and replies, checked by the person who signs them.

Before you reach for computer use, ask whether a simpler route exists. If the system you want to work with has an API, a direct integration is usually faster, cheaper and easier to test than an agent clicking through screens. Our guide to automation vs agents explains how we make that call, and our AI document processing work covers invoice and form extraction where documents, not screens, are the starting point.

The Australian catch: where GPT-6.1 Terra processes your data

As with Gemini 4 Argon earlier this month, the launch coverage skips the data question. OpenAI’s data controls documentation says GPT-6.1 Terra supports only US and EU data residency. OpenAI’s Australian region offers data storage at rest but not regional processing, so even where it applies, inference can happen outside Australia. Regional processing, where available, carries a 10 per cent price premium.

Computer use adds a twist: the screenshots and page text an agent sees are sent to OpenAI as input. OpenAI’s computer use guide notes that turning off storage “does not remove every form of data retention”. If an agent will see customer records, health information or anything your contracts say must stay onshore, check that against the Privacy Act and your client agreements first. Our private LLM options guide covers onshore alternatives.

What Australia’s cyber agency says about AI agents

Terra landed within a day of the Australian Signals Directorate releasing its Annual Cyber Threat Report 2025-26, which called AI “the defining force in the cyber threat landscape over the past year” and warns that “in recent weeks, new threats have emerged from the unintended actions of AI agents and model testing”. OpenAI itself rates Terra as meeting the Critical threshold for cybersecurity capability under its Preparedness Framework, and says it carries the same protections as GPT-6 Astra.

The report’s resilience chapter is clear that businesses should still use AI, but with controls: enforce least privilege access, restrict data exposure to AI systems, validate outputs before use and keep AI-assisted actions auditable. For small businesses, it recommends products that are secure by design, automatic updates and clear responsibility for security monitoring and incident response. ASD’s earlier guidance on AI for small business adds a written AI use policy, vendor due diligence and a human in the loop for legal, medical and financial decisions.

A GPT-6.1 Terra checklist before you automate

1. Start in a sandbox. OpenAI’s guide says to run computer use in an isolated browser or virtual machine with an allow list of sites and actions. Never point a first trial at your live accounting file or banking.
2. Give the agent its own limited login. Create a separate account with only the permissions the task needs, so its actions are logged and can be switched off without touching anyone else’s access.
3. Keep a person on the approve button. OpenAI advises confirming consequential actions such as purchases, data transmission and destructive changes. Payments, supplier bank details and anything sent to customers should always need a human sign-off.
4. Treat what is on screen as untrusted. Text in an email, invoice or web page can try to give the agent new instructions. Prompt injection is one of the risks ASD names, and OpenAI’s guide says page content “cannot grant permission”.
5. Cap and check every run. Set step, time and cost limits, and check the actual result in the system rather than relying on the model’s own report that it finished.
6. Confirm the data path. Decide what the agent may see, where it is processed and how long it is kept, before real customer data goes anywhere near it.

What this means for your automation plans

GPT-6.1 Terra makes screen-based AI agents cheaper and quicker to run, which brings more back-office tasks within reach of a small business budget. The questions that decide whether it saves money are the same as ever: which tasks are worth automating, which systems the agent should touch, who approves its work and where the data goes. If you are weighing this up, our AI agent development, AI automation and LLM integration work covers permissions, approvals, logging and fallbacks. If your team is new to agents, our free AI agents course is a good place to start, and an AI business audit maps where an agent would actually save time.

This is general information based on OpenAI’s announcement and developer documentation, ASD guidance and published media reporting as at 28 October 2026. It is not legal or security advice. Prices, availability and data residency options can change, so check OpenAI’s current documentation and get advice on your own situation before automating work with customer or financial data.

Featured image: Photo by Kindel Media on Pexels.