What OpenAI actually released
First, a naming note, because the headlines are messy. There is no ChatGPT version 6.1. ChatGPT is the app. The models inside it come in tiers, and the current family has GPT-6.1 Sol (the stronger one, out since late September) and now GPT-6.1 Terra (the mid-priced one).
Terra arrived on October 26, according to OpenAI's Codex changelog. OpenAI's safety addendum describes it as a fast, efficient model for high-volume professional work, with strong image analysis and computer use. In plain terms, it is built to read what is on a screen and click, type and move through apps for you.
It succeeds GPT-5.6 Terra. The older model has not been switched off. Its API page still lists it with no retirement notice.
Who can use it and what it costs
Terra shows up in two places inside ChatGPT: Work (the agent mode that builds documents, spreadsheets and reports from your files) and Codex (OpenAI's agent for code and, more and more, general computer tasks). It is not in the regular chat window and it is not on the Free plan. OpenAI's changelog lists Plus, Pro, Business, Enterprise and Edu, though some launch coverage left Business off the list, so check your own account before you plan around it.
For anyone building on the API, the model page lists $1 per million input tokens and $5 per million output tokens. A token is roughly three quarters of a word. The previous Terra was $2 in and $12 out, so this is about half the price. It can also take in a very large amount of text at once (over a million tokens), which matters if you want it to read a long contract or a year of email in one go.
What OpenAI says it is better at
Every number in this section comes from OpenAI's own testing. Nobody independent has checked them yet.
- Fewer mistakes. OpenAI says its error rate on tricky prompts dropped to 16.7%, down from 53.6% for GPT-5.6 Terra, as reported by 9to5Mac.
- Better at multi-step computer work. On a coding and terminal test called Terminal-Bench 4.0, OpenAI reports 50.3% of tasks completed versus 21.5% for the older Terra, per BleepingComputer.
- Faster. OpenAI says Terra finishes tasks more than three times faster on average than its bigger models, also per BleepingComputer.
The Register called it "not a bombshell" but noted it rivals pricier models on some tasks at a fraction of the cost. That is a fair read. This is a value release, not a leap. OpenAI itself still points people to GPT-6.1 Sol for the hardest coding and research work.
The safety fine print
One detail most coverage skipped. In its safety addendum, OpenAI rates Terra "Critical" for cybersecurity capability under its own risk framework, and launched it with the same safeguards as its larger models. The same document says that in an internal test, Terra triggered more prompt injection flags than Sol. Prompt injection means hidden instructions on a web page or in a file that try to trick the AI into doing something you did not ask for.
Our take: a model that clicks around your computer is only as safe as the tasks you hand it. Keep a person approving anything that sends money, emails a customer or touches a password.
What this means for a small business
The interesting part is not the benchmark. It is that the cheaper tier is now good enough to do routine screen work. A few examples of where we would test it:
- Trades and contractors: pull job details from emailed PDFs into your quoting spreadsheet.
- Law and accounting offices: draft a first-pass summary of a long document set before a person reviews it.
- Distributors: copy order details from a supplier portal into your own system when there is no integration.
- Any office: turn a messy folder of notes into a clean weekly report in ChatGPT Work.
Where we would still lean another way: for most small businesses we set up Claude Code as the main AI workhorse, because it is strong at long, careful tasks and easy to keep on a short leash. But if your team already lives in ChatGPT, Terra makes the paid plan more useful than it was last week. We cover OpenAI's tools in our Codex guide.
What we'd do this week
- If you pay for ChatGPT, open Work or Codex and check whether Terra is in your model picker.
- Pick one boring, low-risk task (a report, a data copy job) and time it with and without the AI.
- Do not give it access to banking, payroll or customer email yet.
- If you want help choosing between ChatGPT and Claude for your team, get in touch.