TL;DR
- GPT-6.1 Terra is OpenAI's newest cost-efficient model for everyday agentic work and coding.
- API prices are half or less of GPT-5.6 Terra's: $1 input, $5 output per million tokens.
- OpenAI says it finishes tasks over 3× faster than GPT-6.1 Sol and GPT-6 Astra.
- Independent tests show big terminal-coding gains and fewer made-up answers.
- It trails GPT-6.1 Sol on harder work, so choose Sol or Astra for deep research.
Released on October 26, 2026, OpenAI's GPT-6.1 Terra is built for everyday work. It reads screens and acts across apps. Its API prices are 50% lower than GPT-5.6 Terra.
Pick a model that is too big, and you overpay for routine tasks. Pick one that is too small, and hard tasks fail. Terra aims at the middle ground, and OpenAI is open about where it stops.
This guide explains what GPT-6.1 Terra is and how much it costs. It also covers benchmarks, where to use it, and where it falls short.
What Is GPT-6.1 Terra?
GPT-6.1 Terra is an OpenAI model built for fast, affordable, everyday agentic work. Agentic means it can plan steps and take actions, not just answer questions. Artificial Analysis calls it the first Terra-class model since GPT-5.6. OpenAI positions it as a model for your whole workday.
Where Terra Fits in OpenAI's Lineup
GPT-6.1 Terra sits between OpenAI's heavier and lighter models. OpenAI recommends Sol or GPT-6 Astra for in-depth research and complex coding. Terra targets the everyday work in between. Artificial Analysis ranks it below GPT-6.1 Sol but above GPT-6 Luna.
What Computer Use Means
Computer use means the model can read a screen and act on it. It works across apps, not just inside a chat window. Typical examples include filling in forms or updating spreadsheets. OpenAI says Terra understands what is on screen and takes action across apps.
Key Features of GPT-6.1 Terra
GPT-6.1 Terra focuses on four things: screen understanding, speed, coding, and reliability. Each one targets routine work where cost and speed matter most.
Screen Understanding and Action
OpenAI describes Terra as a model for the whole workday. Here is what that covers:
- Reads documents and application screens
- Takes action across apps
- Writes naturally for emails, summaries, and reports
On Agents' Last Exam, it substantially outperforms GPT-5.6 Terra, according to OpenAI. That benchmark tests how well agents use software at work. It suggests Terra suits workflows built around documents and app screens.
Faster Task Completion
OpenAI says Terra finishes tasks over 3× faster than Astra and Sol on average. It also says Terra matches GPT-6 Astra at one-tenth the price per token. These are OpenAI's own claims, so test them on your workload. Speed matters most when an agent runs many steps.
Coding and Terminal Work
GitHub describes Terra as designed for everyday agentic coding and terminal work. In early testing, it solved noticeably more terminal and C# tasks than GPT-5.6 Terra. GitHub adds that it works in focused, consolidated steps on well-scoped tasks. Artificial Analysis measured a gain of roughly 20 points on Terminal-Bench 4.0.
Fewer Made-Up Answers
Artificial Analysis found Terra guesses far less often when it does not know an answer. In its AA-Omniscience test, the hallucination rate fell from 88% to 38%. Accuracy also rose by 2 points. A 38% guess rate can still be high, so verify important outputs.
GPT-6.1 Terra Pricing
Terra costs $1 per million input tokens and $5 per million output tokens. That is half or less of GPT-5.6 Terra's rates. Tokens are small chunks of text the model reads and writes. OpenAI cites 50% lower standard API prices.
API Price Comparison
| Price per 1M tokens | GPT-5.6 Terra | GPT-6.1 Terra |
|---|---|---|
| Input | $2.00 | $1.00 |
| Output | $12.00 | $5.00 |
| Cache write | $2.50 | $1.25 |
| Cached input | $0.20 | $0.05 |
Cached input now costs one quarter of its old price. Caching helps when you send the same long instructions repeatedly. Check OpenAI's model card for current rates and rate limits.
Cost per Task
Price per token is only part of the story. Artificial Analysis also measured cost per task at max effort. Terra cost $0.40 per task versus $1.40 for GPT-5.6 Terra. It also scored 6 points higher on the Intelligence Index.
Usage Allowance in ChatGPT Work and Codex
Not everyone pays by the token. In ChatGPT Work and Codex, you get a usage allowance with your plan. OpenAI says that allowance goes further with Terra than with GPT-5.6 Terra. It goes far further than with GPT-6.1 Sol or GPT-6 Astra.
For caching, Batch, Flex, and Fast mode rates, see our GPT-6.1 Terra pricing guide.
GPT-6.1 Terra Benchmark Results
Independent tests from Artificial Analysis show Terra gaining intelligence while cutting cost. Its biggest jump is terminal-based coding. Gains elsewhere are smaller.
| Test | GPT-6.1 Terra result | Context |
|---|---|---|
| Intelligence Index (max effort) | +6 points vs GPT-5.6 Terra | 4 below Sol, 10 above Luna |
| Intelligence Index (lower effort) | +7 to +13 points | Larger gains at lower effort |
| Coding Agent Index v2.0 | 48 | Sol scores 54, Luna scores 31 |
| Terminal-Bench 4.0 | About +20 points | Rivals Sol at max effort |
| AA-Omniscience hallucination rate | 38% | Down from 88% |
| GDP.pdf | +5 points | Gains elsewhere are smaller |
Why Effort Settings Matter
Effort settings control how much the model thinks before answering. Artificial Analysis found low and medium effort give the best intelligence per dollar. Other effort levels sit behind that cost-efficiency frontier. A sensible approach is to start low and raise effort only when results disappoint.
How to Read These Numbers
Benchmarks are useful but imperfect. They test set tasks, not your business. OpenAI and Artificial Analysis each use their own tests. Run a small trial on your own work before switching.
For a score-by-score breakdown, see our GPT-6.1 Terra benchmarks guide.
Where to Use GPT-6.1 Terra
You can use Terra in ChatGPT Work, Codex, the OpenAI API, and GitHub Copilot. Access depends on your plan.
| Platform | Who can use it | Notes |
|---|---|---|
| ChatGPT Work and Codex | Plus, Pro, Enterprise, and Edu users | Rolling out now |
| OpenAI API | Developers | Model name: gpt-6.1-terra |
| GitHub Copilot | Pro, Pro+, Max, Business, and Enterprise | Generally available, gradual rollout |
ChatGPT Work and Codex
OpenAI says Terra is rolling out now to all Plus, Pro, Enterprise, and Edu users. It is available in both ChatGPT Work and Codex. For deep research or complex coding, OpenAI suggests picking Sol or Astra instead.
OpenAI API
Developers can call the model as gpt-6.1-terra. OpenAI's model card lists pricing, rate limits, and extra details. Review it before you plan your budget.
GitHub Copilot
GitHub made Terra generally available in Copilot on October 26, 2026. You can pick it in the model picker across these surfaces:
- Visual Studio Code and Visual Studio
- JetBrains IDEs, Xcode, and Eclipse
- Copilot CLI, the Copilot cloud agent, and the Copilot app
- github.com and GitHub Mobile on iOS and Android
Rollout is gradual, so check back if you do not see it yet. Business and Enterprise administrators control access through the model policy in Copilot settings. GitHub bills the model at provider list pricing under usage-based billing.
Who Should Use GPT-6.1 Terra?
Terra suits teams that run many routine tasks and care about cost. It is less suited to work where one mistake is expensive.
Good Fits
- High-volume agent tasks where per-token costs add up
- Workflows involving documents, forms, and app screens
- Everyday coding and terminal work with clear scope
- Drafting emails, summaries, and reports
When to Choose Sol or Astra
OpenAI recommends GPT-6.1 Sol or GPT-6 Astra for in-depth research or complex coding. Independent scores agree. Terra sits 4 points below Sol on intelligence and 6 points below on coding agents. If one mistake is costly, pay for the stronger model. Otherwise, start with Terra and move up only when needed.
For a side-by-side of the two tiers, see our GPT-6.1 Terra vs GPT-6.1 Sol comparison.
Limitations to Keep in Mind
GPT-6.1 Terra is strong for its price, but it is not perfect. Keep these limits in mind:
- Not the strongest model: It trails Sol on both the Intelligence Index and the Coding Agent Index.
- Still guesses: A 38% hallucination rate means important outputs need checking.
- Vendor claims: The speed figures and Agents' Last Exam results come from OpenAI itself.
- Scoped tasks work best: GitHub notes its strengths on well-scoped tasks, not open-ended ones.
- Computer use needs oversight: Review any action involving money, deletions, or customer messages.
- Availability varies: Rollout is gradual, and access depends on your plan.
None of these are dealbreakers. They are reasons to test before you commit.
Conclusion
GPT-6.1 Terra shows that efficient models can now handle serious everyday work. It is not the most powerful option, and OpenAI says so. But for routine agentic tasks, it offers a strong mix of speed, cost, and capability.
Models like GPT-6.1 Terra are powerful building blocks, but a model alone is not a product. Emergent is an AI-powered vibe coding platform for building functional, deployable apps. Describe your idea in plain language and start building with Emergent today.
Most AI app builders stop at prototypes. Emergent creates production-ready apps you can actually launch.
- Production-ready apps
- Web & mobile apps
- Deploy in minutes
