
Breakdown / AI prices · Computer use
GPT-6.1 Terra costs less per token. What does that buy?
OpenAI’s new model is cheaper to run through its API. For an invoice or a coding job, the saving depends on how much work it takes to get a correct result.
OpenAI has launched GPT-6.1 Terra, an AI model for reading images, writing and working through tasks on a computer. Its standard API prices are half those of GPT-5.6 Terra and GPT-6.1 Sol, according to the company. Launch announcement
Say you want an assistant to enter an invoice into your accounting software. It needs to read the document, find the supplier and amounts, fill the right fields and check what it saved. Terra can help interpret the document and decide what to do. The app around it still needs to provide the tools and permission to use them. We haven’t tested that workflow with Terra.
A lower price could give you more room to try this kind of automation. How much it saves depends on how often someone has to step in and fix the work.
What it costs, and who gets it
OpenAI says Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and its coding assistant, Codex. Developers can also use the API, which lets their own software call the model. The API rates aren’t the price of a ChatGPT subscription. Availability
The standard rates, in US dollars, are $1 per million input tokens and $5 per million output tokens. Tokens are the small pieces of text a model processes. Input is what you send it; output is what it generates. Model documentation
A request with 100,000 input tokens and 10,000 output tokens would cost 15 cents at those rates, before tool fees or other charges. That’s a price calculation, rather than the cost of an invoice we’ve processed.
There are conditions. Send more than 272,000 input tokens in one request and the higher rates apply to the whole request: twice the input and cache rates, and 1.5 times the output rate. Fast mode also costs more. Caching, which reuses material the model has already processed, can cut input charges, but storing that material has a separate cost. Pricing details
How much time should it spend thinking?
You can let Terra spend more processing on an answer. OpenAI calls this setting reasoning effort. Turning it up can improve a result, but it can also use more tokens and cost more. Effort settings
OpenAI says that at medium effort, Terra scores 15 percentage points above GPT-5.6 Terra on Agents’ Last Exam, at about one-third the cost. That test includes professional tasks such as turning architectural drawings into a 3D model. OpenAI’s results
Artificial Analysis, an outside company that tests AI models, shows why the setting matters. When we checked on October 28, Terra scored 29 at low effort and 51 at max in its combined tests, where higher is better. Its calculated average cost per test task rose from about 7 cents to 96 cents across those settings. Those are figures from its selected tests, so they won’t tell you what your invoice job will cost. Evaluator’s results
It also reports output speed of roughly 91 tokens a second. An assistant may still spend time waiting for an app to load, calling tools or retrying a failed step. The time it takes to produce text tells you only part of the story. Speed results
We haven’t compared Terra’s finished work or actual bills with another model. For that invoice job, we’d count correct entries, duplicates and the time spent fixing mistakes. A cheaper model earns its place when it leaves you with less paperwork to do, including the cleanup.
Sources
- OpenAI’s Terra releaseOpenAI
- API rates and effort settingsOpenAI
- Outside Terra evaluationArtificial Analysis
- Evaluator methodsArtificial Analysis