GPT-6.1 Terra: The Everyday Workhorse in OpenAI’s Line-up

GPT-6.1 Terra is OpenAI’s mid-sized model, launched on 26 October 2026. It costs $1 per million input tokens and $5 per million output tokens, which is half the price of GPT-6.1 Sol. It supports about 1 million tokens of context, image input and computer use. It’s best suited to high-volume everyday work, with Sol kept for the hardest tasks.

Key takeaways

Half the price of GPT-6.1 Sol, with batch and caching discounts on top
Computer use lets an agent operate software that has no API
Independent testing places it slightly below Sol, which fits its role
Keep a human approval step for any action that submits or saves data

Where Terra fits

Think of a delivery fleet. You wouldn’t send a truck to deliver one parcel. OpenAI’s current line-up works the same way: Luna is the small, cheap option, Sol is the heavy-duty model, and Terra is the van that handles most daily jobs.

What you get

Price: $1 input and $5 output per million tokens (up to 272K input). Cached input costs $0.05. Batch and Flex processing cost $0.50 input and $2.50 output.
Context: about 1.05 million tokens in total (922K input, 128K output).
Inputs: text and images.
Tools: web search, file search, code interpreter, MCP and computer use.
Effort: adjustable from none to max, with medium as the default.
Where: ChatGPT Work and Codex (Plus, Pro, Business, Enterprise, Edu), and the API as gpt-6.1-terra.

A practical example

An accounting firm in Sydney receives hundreds of supplier invoices by email each month. Its old desktop accounting software has no API, so staff type every invoice in by hand. With Terra’s computer use, an agent can read each invoice PDF, open the accounting app, enter the supplier, amount and date, and then stop for a person to approve before saving. Running the job overnight in batch mode lowers the cost further. 

How to start, step by step

Choose Terra in ChatGPT Work, in Codex (codex -m gpt-6.1-terra, CLI 0.159.0+), or in the API.
Begin at medium effort and raise it only for tricky cases.
Use batch mode for large, non-urgent jobs.
Keep standard instructions identical across runs so cached input stays cheap.
Add a human approval step before anything is submitted or saved.

What’s verified and what isn’t

Most published benchmarks are OpenAI’s own. Independent scoring from Artificial Analysis gives Terra 48 at max effort, against 52 for GPT-6.1 Sol. That fits its role as the value option, not the top model.

Frequently asked questions

How much does GPT-6.1 Terra cost?

$1 per million input tokens and $5 per million output tokens, half the price of GPT-6.1 Sol. Batch processing halves it again.

Can GPT-6.1 Terra use software on a computer?

Yes. It supports computer use, so an agent can click, type and fill forms in apps, ideally with a human approving final actions.

Is Terra better than GPT-6.1 Sol?

No. Independent tests place it slightly below Sol. It’s designed for high-volume everyday work at a lower price.

Where can I use GPT-6.1 Terra?

In ChatGPT Work and Codex on paid plans, and through the API as gpt-6.1-terra.