OpenAI's GPT-6.1 Terra Works Inside Your Apps at Half the Price of Its Predecessor

The new model understands what is on screen and takes action across apps. OpenAI says it matches its flagship GPT-6 Astra on a workplace agent test at a tenth of the cost per token.

SAN FRANCISCO - OpenAI on Monday released GPT-6.1 Terra, a faster and cheaper model built to operate computers on a user’s behalf, reading what is on the screen and taking actions across apps to finish routine office work. The company said in its announcement that the model is aimed at “the communication and computer-based tasks that fill the workday, from drafting proposals and interpreting technical drawings to filling out forms and editing videos.”

Terra succeeds GPT-5.6 Terra and is meant to balance speed, capability and cost, while GPT-6 Astra remains what OpenAI calls its most capable frontier model. Terra’s standard API prices are 50 percent lower than those of GPT-5.6 Terra and GPT-6.1 Sol: $1 per million input tokens, 5 cents per million cached input tokens and $5 per million output tokens.

OpenAI announced the model on X on Monday. Screenshot: @OpenAI on X

The headline claim is about agents. On Agents’ Last Exam, a benchmark of professional tasks such as turning architectural drawings into a 3D model, OpenAI said GPT-6.1 Terra at medium reasoning effort scores 15 percentage points higher than GPT-5.6 Terra at about a third of the cost, and performs on par with GPT-6 Astra at a tenth of its price per token. At maximum effort, the company said, Terra is more than three times as token efficient as Anthropic’s Claude Opus 5.5 and a quarter of its price per token, for roughly nine times lower cost per task. OpenAI also said Terra finishes tasks more than three times faster than both Astra and GPT-6.1 Sol.

OpenAI’s chart of Agents’ Last Exam scores against cost per task. Image: OpenAI

OpenAI is also pitching the model as more accurate. On its internal benchmark of prompts designed to provoke errors, the share of answers with a factual mistake at low reasoning effort fell from 53.6 percent with GPT-5.6 Terra to 16.7 percent with GPT-6.1 Terra, the company said. At maximum effort the rate fell to 4.3 percent, close to Astra and GPT-6.1 Sol. On Terminal-Bench 4.0, a test of coding agents working from the command line, Terra solved 50.3 percent of tasks at maximum effort, more than double GPT-5.6 Terra’s published 21.5 percent, though OpenAI said Sol and Astra remain the better choice for the hardest coding work.

OpenAI’s chart of factual error rates on difficult prompts, where lower is better. Image: OpenAI

The company said GPT-6.1 Terra meets the “Critical” threshold in cybersecurity under its Preparedness Framework and keeps the protections it built for GPT-6 Astra. In OpenAI’s prompt-injection tests, the model showed strong resistance to malicious instructions meant to pull it away from a user’s task, OpenAI said, with more detail in an addendum to the GPT-6 Astra system card.

Video: OpenAI / Vimeo

GPT-6.1 Terra is rolling out now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, OpenAI said on X, and developers can call it through the API as gpt-6.1-terra. The company said the same ChatGPT usage allowance covers more tasks with Terra than with GPT-5.6 Terra, and substantially more than with GPT-6.1 Sol or GPT-6 Astra.