OpenAI released GPT-6.1 Terra on Oct. 26, a mid-tier model aimed at computer use and visual reasoning, with API prices it says are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol.
Computer use means the model operates software on a screen the way a person would, by reading what’s displayed, clicking and typing. OpenAI described Terra as “a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing,” The Verge’s Jay Peters reported. The company said it “excels at the communication and computer-based tasks that fill the workday.”
Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, OpenAI’s coding agent. Developers can call it through the API as gpt-6.1-terra.
Price and access
Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens, according to OpenAI. Tokens are the small chunks of text a model reads and writes, and cached input is text the model has already processed recently, which costs less to reuse.
OpenAI’s developer documentation lists a context window of 1.05 million tokens, the amount of text the model can consider at once. It accepts text and images and offers reasoning-effort settings that range from none to max.
Amazon made the model generally available on Amazon Bedrock, its service for running AI models, the same day. AWS said customers can call Terra through Bedrock or set ChatGPT Work and Codex to use the model there. It suggested uses such as drafting proposals, reading technical drawings and investigating codebases.
How it scores
OpenAI’s headline result is on Agents’ Last Exam, a benchmark of professional tasks done on a computer. Terra scores 15 percentage points higher than GPT-5.6 Terra there at about one-third the cost, OpenAI says.
AWS’s announcement added two details from OpenAI’s evaluation. It said Terra performed on par with GPT-6 Astra, OpenAI’s larger model, on that test and finished tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol.
On Terminal-Bench 4.0, a test of coding in a command-line terminal, Terra solved 50.3% of tasks at maximum effort, compared with 21.5% for GPT-5.6 Terra, OpenAI says. The company still recommends its larger models for the most complex coding.
Outside tests put it lower in some places. Replit’s ViBench 2.0, released two days later, found that Terra passed every test in 47% of its full app builds, at about half the cost of GPT-6.1 Sol, which passed 53% on the same public products.
OpenAI also reports fewer mistakes. On its internal set of prompts designed to trigger errors, Terra’s error rate at low reasoning effort fell from 53.6% for the previous Terra to 16.7%. These are OpenAI’s own figures.
On safety, OpenAI says Terra meets the Critical threshold for cybersecurity under its Preparedness Framework, the company’s system for rating a model’s potential for serious harm. It ships with the same protections OpenAI built for GPT-6 Astra, according to the company, and is covered in an addendum to the Astra system card.
Source: OpenAI, “Introducing GPT-6.1 Terra”
Additional sources: The Verge, Jay Peters, “GPT-6.1 Terra is very good at computer use, OpenAI says.”; AWS, “OpenAI GPT-6.1 Terra is now generally available on Amazon Bedrock”; OpenAI API docs, OpenAI, “GPT-6.1 Terra”
