OpenAI Unveils GPT-6.1 Terra, Cutting Costs and Improving Performance

Overview

• OpenAI has launched GPT-6.1 Terra on October 26, offering a 50% reduction in standard API prices compared to GPT-5.6 Terra.
• Available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, as well as developers.
• Internal tests show a significant decrease in factual errors and improved task completion rates for Terra.

On October 26, OpenAI introduced GPT-6.1 Terra, a new version of its AI model designed to provide faster and more accurate performance at half the cost of its predecessor, GPT-5.6 Terra. This model is particularly aimed at routine office tasks, enhancing usability by allowing the AI to interact with screens and applications like a human would.

The 50% price reduction applies to the standard API fees that developers incur when integrating the model into their applications. OpenAI emphasized that GPT-6.1 Terra represents a notable advancement over the previous version, stating, “It’s a significant step up over GPT-5.6 Terra, with 50% lower standard API prices.”

Terra is designed for everyday tasks, functioning within OpenAI's ChatGPT Work agent mode, which can navigate applications and complete online tasks autonomously. For users in the free tier, the older GPT-5.6 Luna remains available.

In internal evaluations, the model displayed a substantial drop in factual inaccuracies, with errors at low reasoning effort decreasing from 53.6% for GPT-5.6 Terra to just 16.7% for Terra. At maximum reasoning effort, the error rate also improved significantly, showcasing Terra's enhanced reliability.

Furthermore, on the Terminal-Bench 4.0 test, which assesses AI performance in executing command-line tasks, Terra achieved a successful task completion rate of 50.3%, a notable improvement from GPT-5.6 Terra's 21.5%. Additionally, OpenAI reports that Terra completes tasks over three times faster than its predecessors, GPT-6 Astra and GPT-6.1 Sol, further underscoring its efficiency.

OpenAI first introduced the Terra model on June 26 as part of the GPT-5.6 family, following a limited preview requested by the Trump administration, and made it publicly available on July 9. While GPT-6.1 Terra is optimized for everyday use, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra for more complex coding tasks or in-depth research.