· Sam Fielding

OpenAI's GPT-6.1 Terra Halves Its API Price: What To Ask About Your Automations

On 26 October 2026, OpenAI released GPT-6.1 Terra at half the API price of the previous Terra. For Australian businesses using AI to read invoices, sort email or enter data, the saving only lands once someone switches the model and tests it on real work.

On 26 October 2026, OpenAI released GPT-6.1 Terra, a model built for visual reasoning, computer use and writing. Developers can call it through the OpenAI API as gpt-6.1-terra. ChatGPT users on paid plans get it inside Work and Codex.

What OpenAI Released

OpenAI’s standard API price for Terra is USD 1 per million input tokens, USD 0.05 per million cached input tokens and USD 5 per million output tokens. Tokens are the small chunks of text a model reads and writes, and API usage is billed by them. OpenAI says the new price is 50% lower than both GPT-5.6 Terra, the previous version, and GPT-6.1 Sol.

OpenAI’s post also contains a bigger number. It says Terra performs on par with GPT-6 Astra on its Agents’ Last Exam benchmark at one-tenth of Astra’s price per token. Astra is OpenAI’s most capable model, so that figure compares Terra with the top of the range. Against the previous Terra, the cut is 50%.

OpenAI also says Terra completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. OpenAI ran these benchmarks itself.

For ChatGPT, OpenAI’s release notes say Terra is rolling out in Work and Codex on Plus, Pro, Business, Enterprise and Edu plans, subject to your plan, region and workspace settings. Overall usage limits are unchanged, and OpenAI says the same allowance covers more tasks with Terra than with GPT-5.6 Terra. On 29 October 2026, OpenAI doubled the Work and Codex allowance on its Pro 100 and Pro 500 plans and cancelled a planned allowance cut on Pro 200. The release notes list no change to ChatGPT plan prices.

What It Means For An Australian Business Running Automations

OpenAI picked an example most operators will recognise. It says the lower price makes it “practical for developers to automate high-volume work, like reading incoming invoices and entering their details into accounting software.”

That is where model price matters most. An automation that reads every supplier bill or sorts every inbound email calls the model once per item. Each call costs very little, but it repeats hundreds or thousands of times a month. Halve the price per token and, for the same token count, that line of the bill halves.

Four things stand between that headline and your actual bill.

Nothing switches by itself. Most automations call a model by its exact name. Until someone changes that setting and tests the result, you keep paying the old rate.

Cost per job is the number that counts. What you pay depends on how many tokens each job uses, and the reasoning effort setting changes that. OpenAI’s own numbers show the gap: at medium effort, it says Terra scored 15 percentage points higher than GPT-5.6 Terra on Agents’ Last Exam at about one-third of the cost. Ask for cost per invoice or per email, and ignore cost per million tokens.

Cheaper models still make mistakes. On OpenAI’s internal benchmark of prompts designed to cause errors, Terra’s error rate was 16.7% at low reasoning effort and 4.3% at maximum effort. If an automation writes supplier bills into Xero, a person should still review anything it flags as uncertain before it posts.

Some jobs run on a bigger model than they need. If your inbox triage or data entry runs on GPT-6 Astra or GPT-6.1 Sol, OpenAI’s figures put Terra at more than three times the speed and between half and one-tenth of the price per token. OpenAI says Sol and Astra remain the better choice for the most complex coding tasks. Sorting email is nowhere near that.

If your automations run on a different provider, the same questions apply.

What To Ask Whoever Builds Your Automations

Whether that is an agency or someone on your team, these questions get a straight answer:

  • Which AI model does each automation call, and from which provider?
  • What does one run cost: one invoice read, one email sorted?
  • How is accuracy checked, and what goes to a person for review?
  • Who holds the AI usage account, and can I see the bill?

The Next Step

Send those questions to whoever built your automations. Then ask them to run GPT-6.1 Terra side by side with your current model on a month of your real invoices or emails, and report two numbers back: cost per item and error rate. If both hold up, switching is usually a small change. If they cannot tell you which model you are on now, fix that first.

For a fuller view of what an automation should cost to run each month, read What Ongoing Automation Maintenance Actually Costs.

Sources

Go Deeper

Sam Fielding
Sam Fielding
Managing Director, Echelon AI Solutions

Sam co-founded Echelon AI Solutions and leads transformation strategy, client engagements and growth. He has built and operated businesses across marketing and AI education, and has guided companies in retail, trades, hospitality and professional services through operational change. His focus is making AI earn its place through measurable business performance.

All AI news