Last week OpenAI held back a model because it wasn't honest enough. This week another one ships. No contradiction: two different models with similar-sounding names. Welcome to the AI world.
What's new?
On October 26, OpenAI launched GPT-6.1 Terra. According to reports, it runs in ChatGPT and in Codex, OpenAI's tool for programming.
The promise: fast and capable at computer tasks, at a lower price. OpenAI says Terra understands what is on screen and then operates apps itself. That is called computer use (the AI clicks and types instead of just producing text).
For context: Terra is not the flagship. OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra for deep research and hard coding. Terra is more the everyday workhorse.
What does it cost?
For developers, the current picture: through the API (the direct access for programs), Terra costs, per OpenAI, $1 per million input tokens and $5 per million output tokens. A token is roughly a syllable. By OpenAI's account that is half of what the predecessor GPT-5.6 Terra costs.
In ChatGPT itself, Terra is currently available to Plus, Pro, Enterprise and Edu users, in the ChatGPT Work area. We found nothing on whether or when free users get it. Free tiers are tightly limited at every big provider anyway: fewer messages, weaker models. That isn't spite. Compute costs real money.
Is it for me?
Beginners without a paid plan: nothing to do yet. Keep trying what your free access offers.
With a Plus subscription: check whether Terra is selectable in ChatGPT. OpenAI says your allowance stretches further than with the predecessor. Test that yourself.
Developers: this is where the appeal is. Half the price for a model that can read screens makes agent experiments cheaper.
One honest caveat: that Terra is "fast and capable" is, for now, the vendor's claim. Independent tests are still thin.
What you can try
- Small task, clear limit. Let the AI sort a spreadsheet, not operate your bank account.
- Watch along. With computer use you see every step. Use that.
- Ask back. "What exactly did you do?" Asking that after every task exposes weaknesses early. Why it matters is in our article on the halted GPT-6.1 Astra.
For the predecessor flagship, see our GPT-6 Astra article.
Bottom line
Terra is no quantum leap, but a price-tag update with a view of the screen. That is exactly what makes it interesting: when AI agents get cheaper, more people try them. Good and slightly risky at once. Start small, watch, and only trust it after it has worked several times.