World 00:44 UTC
EN – Choose language: English

Choose language

Back to overview

Technology

OpenAI Launches GPT-6.1 Terra: Computer-Control AI Model at Half the Price

GPT-6.1 Terra is designed to operate programs on screen on its own. Plus, Pro, Enterprise and Edu subscribers get it immediately in ChatGPT Work and Codex.

By

Automatically translated from German · may contain errors · Read the original

Old town alley with red-tiled roofs and a yellow streetcar; at left, an input window with an English prompt for a 3D model
OpenAI unveils GPT-6.1 Terra: The model operates apps on its own, costs half as much as its predecessor in the API and is available now.

OpenAI released the AI model GPT-6.1 Terra on Monday, Oct. 26. It is designed to capture on-screen content, carry out actions across multiple applications and write in a natural style. Standard prices through the application programming interface (API) are 50 percent below those of its predecessor, GPT-5.6 Terra.

According to the company, the model is available immediately to subscribers of the Plus, Pro, Enterprise and Edu plans in ChatGPT Work and in the coding tool Codex. Developers can access it through the API under the name gpt-6.1-terra. On X, OpenAI described it as a "significant step forward" from GPT-5.6 Terra. The usage allowance in ChatGPT Work and Codex stretches further with the new model than with its predecessor, and significantly further than with GPT-6.1 Sol or GPT-6 Astra, the company said.

X @OpenAI

Showing the post loads it from X, which transfers data to X.

View on X ↗

One dollar per million input tokens

According to the model page for developers, one million input tokens cost $1, cached input costs $0.05, and one million output tokens cost $5. Tokens are the units of text in which language models process and bill inputs and outputs.

OpenAI describes gpt-6.1-terra there as a fast, efficient model for visual reasoning, precise computer use and natural writing. Computer use means that the AI operates programs on its own, for example by clicking and typing.

Vendor tests and independent measurements

In OpenAI's own tests, Terra clearly outperforms its predecessor on the Agents’ Last Exam benchmark. At medium compute effort, Terra is 15 percentage points ahead of its predecessor, at about one-third of the cost, according to OpenAI. These results have not been independently verified.

According to OpenAI, Terra reaches the level of GPT-6 Astra on the test but costs only one-tenth as much per token. The company said the model completes tasks on average more than three times as fast as Astra and GPT-6.1 Sol.

The independent benchmark provider Artificial Analysis measures a gain of six points on its Intelligence Index over GPT-5.6 Terra at maximum compute effort. The cost per task falls from $1.40 to $0.40, it said. In the provider's rating, Terra trails GPT-6.1 Sol by four points and leads GPT-6 Luna by ten points.

The hallucination rate, meaning the share of false, fabricated statements, fell from 88 to 38 percent on the AA-Omniscience test, according to Artificial Analysis. On the Coding Agent Index, Terra scores 48 points and GPT-6.1 Sol scores 54.

Terra returns, Astra update scrapped

When it launched GPT-6 in September, OpenAI left out the Terra line, according to 9to5Mac. Also in September, the company scrapped the model GPT-6.1 Astra after it failed internal safety tests, according to The Next Web.

According to OpenAI, Terra itself reaches the "Critical" level for cybersecurity under the company's Preparedness Framework. The company uses this framework to assess the risks of its models.

More stories