Key Takeaways
- GPT-6.1 Terra is an OpenAI model focused on computer use: it reads the screen, clicks and types across apps, and writes naturally for everyday work.
- It is available through ChatGPT Work, Codex and the OpenAI API under the model ID gpt-6.1-terra, with API pricing OpenAI says is 50% lower than GPT-5.6 Terra.
- OpenAI claims it matches GPT-6 Astra on the Agents' Last Exam benchmark at roughly one-tenth the cost per token; treat that as a vendor figure until you test it yourself.
- The lower per-token price matters most for multi-step agents, because a single task can involve many screen-reading and clicking steps.
Quick Facts
| Type | Computer-use AI model with API access |
|---|---|
| Devices | ChatGPT Work, Codex and the OpenAI API |
| Price | API price 50% lower than GPT-5.6 Terra, per OpenAI |
| Languages | Not specified; check the official site |
| Highlight | Screen understanding and action across apps |
| Learning curve | Easy in ChatGPT Work; moderate for API agents |
| Category | AI & Models |
| Creator | OpenAI |
| Official site | openai.com ↗ |
| Heat | 🔥 3.7M Heat |
| Last updated |
What GPT-6.1 Terra Actually Is
GPT-6.1 Terra is a new model from OpenAI, announced on October 26, 2026, and it belongs to a family of models built around computer use rather than chat alone. In OpenAI's own description, it understands what is on screen, takes action across apps, writes naturally, and is meant to be a more capable model for your whole workday with the efficiency to take on more of it. That combination of screen awareness and action is the core of the pitch.
The name matters less than the positioning. Terra sits below OpenAI's top-tier Astra line on cost, and the company says the API price is 50% lower than GPT-5.6 Terra, the previous Terra model. For anyone running agents that read a screen, click, type and check results, a lower price per token directly changes how many tasks are worth automating, because each step in a task consumes tokens.
This is not a standalone app with its own landing page. Access comes through three places: ChatGPT Work, Codex and the OpenAI API, where the model ID is gpt-6.1-terra. If you are looking for a dedicated product page for this title, there is not one yet; the creator's official site is openai.com, and that is where plan details and full pricing tables live.
- Built for the workday, not just chat
OpenAI frames Terra as a model for your whole workday, meaning it is designed to handle routine multi-app chores rather than only answering questions.
- Three access routes
You can reach it through ChatGPT Work, Codex and the API under the model ID gpt-6.1-terra, so both non-developers and engineering teams have a path in.
- The price is the headline
A 50% lower API price than GPT-5.6 Terra is the number OpenAI leads with, and it is the figure most likely to change your build-versus-buy math.
How Computer Use Works Step by Step
Computer-use models operate in a loop rather than a single pass. The model looks at what is on screen, decides on an action, performs it by clicking or typing, then observes the result before deciding what to do next. A single task such as filling a form across two apps might involve dozens of these cycles. Because every cycle sends screen context and receives a response, token consumption adds up quickly, which is why per-token price is such a big deal for this category.
Terra's stated strengths map onto that loop. Understanding what is on screen is the perception step, taking action across apps is the execution step, and writing naturally covers the output you actually read, such as summaries, emails or notes. OpenAI's claim is that GPT-6.1 Terra matches GPT-6 Astra on the Agents' Last Exam benchmark while costing about one-tenth as much per token. That is a vendor figure, not an independent result, so treat it as a starting hypothesis rather than a settled fact.
The practical implication is that Terra is aimed at teams who want to automate repetitive, multi-app work without paying top-tier prices for every step. If your agent spends most of its time reading screens and clicking buttons, a cheaper model with comparable benchmark performance changes the economics. If your task needs deep reasoning rather than screen control, a different model may still be the better fit.
- Perception, action, verification
The loop is read the screen, act, check the result. Failures usually show up in the verification step, so build in a way to confirm each action landed.
- Token cost compounds
A long task can mean many model calls. Halving the per-token price roughly halves the cost of that task, which is why the pricing change matters more than it looks.
- Benchmark claims need your own test
OpenAI says Terra matches GPT-6 Astra on Agents' Last Exam at about one-tenth the cost per token. Run your own tasks before trusting that for production.
Getting Started With Terra in ChatGPT Work, Codex and the API
The easiest entry point is ChatGPT Work, where you can hand off routine chores and see how the model handles multi-app tasks without writing code. Start with something low-risk and repetitive, such as collecting information from a few screens or drafting a summary from what is open. Watch where it hesitates, because hesitation usually signals a step that needs a clearer instruction or a confirmation check.
For developers, the API route is where Terra's price advantage shows up. You call the model under the ID gpt-6.1-terra, and OpenAI says the API price is 50% lower than GPT-5.6 Terra. Before committing, check the official site for current pricing tables, rate limits and plan details, since those are the numbers that decide whether a workflow is viable at scale. Codex sits between the two, useful if your work is already code-adjacent.
Whichever route you take, design for verification. Computer-use agents can misread a screen or click the wrong element, so a cheap model that runs many steps needs guardrails: confirmations before destructive actions, logging of each step, and a human review point for anything irreversible. The lower price makes more automation affordable, but it does not remove the need for oversight.
- Start with a low-stakes chore
Pick a repetitive task where a mistake is easy to spot and undo. That gives you a clean read on Terra's reliability before you trust it with anything important.
- Check the official pricing tables
The 50% figure is OpenAI's comparison to GPT-5.6 Terra. Actual rates, limits and plan details sit on the official site, openai.com, and can change.
- Add confirmations and logs
Because agents act across apps, log every step and require confirmation before irreversible actions. Cheap steps still need a safety net.
How Terra Compares and Who Should Use It
The natural comparison is inside OpenAI's own lineup. GPT-6 Astra is positioned as the more capable, more expensive option, and OpenAI claims Terra matches it on the Agents' Last Exam benchmark at roughly one-tenth the cost per token. GPT-5.6 Terra is the previous Terra model, and the new release undercuts it by 50% on API price. That gives developers a clear trade-off: reach for Astra when a task demands maximum capability, and reach for Terra when you are running many steps and cost per task is the constraint.
Against the broader field of computer-use and agent models, Terra's differentiator is the combination of screen control and price rather than any single breakthrough. Many models can read a screen; fewer are pitched as cheap enough to run all day. That makes Terra most relevant to teams building agents on the API and to ChatGPT Work and Codex users who hand off routine, multi-app chores during the day.
The limitations are worth stating plainly. The benchmark comparison is OpenAI's own claim, not an independent measurement. Language support, exact rate limits and plan availability are not fully specified in the announcement, so check the official site before you plan around them. And computer use remains a category where reliability varies by task, so pilot before you scale. If your work is mostly text generation with occasional screen reading, Terra may be more than you need.
- Terra versus Astra
Astra is the top-tier option; Terra is the cheaper workhorse. OpenAI says Terra matches Astra on Agents' Last Exam at about one-tenth the cost per token.
- Terra versus GPT-5.6 Terra
The new model is 50% cheaper on the API than the previous Terra, so upgrading is mostly a cost decision rather than a capability leap.
- Who it is for
Agent builders on the API and ChatGPT Work or Codex users with repetitive multi-app chores get the most from it. Deep-reasoning tasks may still favor a larger model.
Step-by-Step Guide
Pick your access route
Decide whether you will use ChatGPT Work, Codex or the API. ChatGPT Work is best for hands-on tasks without code, Codex suits code-adjacent work, and the API is for building agents. All three reach the same model.
Confirm current pricing and limits
OpenAI says the API price is 50% lower than GPT-5.6 Terra, but exact rates and limits live on the official site, openai.com. Check those tables before you commit a workflow to the model.
Choose a repetitive, low-risk task
Start with something you do often and can easily verify, such as gathering details from several screens or drafting a summary. Avoid irreversible actions for your first run so mistakes stay cheap.
Run it and watch each step
Let the model read the screen, act and check results, and note where it hesitates or misreads. Those spots usually need clearer instructions or a confirmation step before the next action.
Add guardrails and scale up
Once the task works reliably, add logging and confirmations before destructive actions, then expand to more workflows. Re-test after any model or pricing change on the official site.
Frequently Asked Questions
What is GPT-6.1 Terra?
GPT-6.1 Terra is an OpenAI model announced on October 26, 2026, built around computer use. According to OpenAI, it understands what is on screen, takes action across apps and writes naturally, positioning it as a capable model for your whole workday. It is available through ChatGPT Work, Codex and the OpenAI API under the model ID gpt-6.1-terra.
How much does GPT-6.1 Terra cost?
OpenAI says the API price is 50% lower than GPT-5.6 Terra, the previous Terra model. That is the headline figure, but exact rates, plan details and limits are published in OpenAI's official materials, so check the official site, openai.com, for current numbers before budgeting. ChatGPT Work and Codex access may follow separate plan terms.
How does Terra compare with GPT-6 Astra?
OpenAI claims GPT-6.1 Terra matches GPT-6 Astra on the Agents' Last Exam benchmark while costing about one-tenth as much per token. Treat that as the vendor's own figure rather than an independent result. In practice, Astra is positioned as the more capable, more expensive option, while Terra is the cheaper workhorse for high-volume, multi-step agent tasks.
Where can I use GPT-6.1 Terra?
Access comes through three routes: ChatGPT Work, Codex and the OpenAI API, where the model ID is gpt-6.1-terra. There is no dedicated product page for this title yet, so the creator's official site, openai.com, is the place to confirm availability, pricing and plan details. Availability may vary by plan and region.
Is GPT-6.1 Terra good for building agents?
It is aimed squarely at that use case. Computer-use agents run many steps per task, reading screens, clicking and typing, and each step costs tokens, so a lower per-token price directly changes how many tasks are worth automating. That makes Terra most relevant to teams building agents on the API, though you should still pilot it on your own tasks before scaling.
