▲ OpenAI
OpenAI has unveiled "GPT-6.1 Terra," an artificial intelligence (AI) model capable of performing a wide range of tasks from document writing and technical drawing interpretation to computer operations.
The new model is characterized by enhanced capabilities in understanding visual information and handling complex tasks compared to previous models, while lowering usage fees by half to reduce the burden on businesses and developers utilizing AI for their work.
OpenAI announced GPT-6.1 Terra today (October 27), which supports routine administrative and professional tasks such as drafting proposals, analyzing technical blueprints, entering forms, and editing videos.
This model is designed to efficiently handle multi-step tasks by combining reasoning capabilities that analyze visual information, such as images and documents, with writing and computer execution functions.
In particular, it has lowered standard application programming interface (API) usage fees by 50% compared to the previous "GPT-5.6 Terra" and "GPT-6.1 Sol," helping to reduce cost burdens when automating large volumes of work with AI.
It also showed improved performance over previous models in evaluations measuring actual task execution capabilities.
In the "Agents' Last Exam," which evaluates the performance of various specialized tasks, it scored 15 percentage points higher than GPT-5.6 Terra when using a medium level of reasoning.
It also outperformed GPT-5.6 Terra in evaluations assessing the understanding and judgment of information within images and completing tasks through multiple steps.
▲ OpenAI's 'GPT-6.1 Terra' scored higher than its predecessor, GPT-5.6 Terra, in 'Agents' Last Exam v1'—which evaluates professional task performance—while lowering the cost per task.
Factual accuracy has also improved.
In an internal OpenAI evaluation consisting of questions designed to elicit errors, applying a low reasoning level resulted in an error rate of 16.7% for GPT-6.1 Terra.
This is about 37 percentage points lower than the 53.6% error rate of GPT-5.6 Terra.
In "Terminal-Bench 4.0," which evaluates coding capabilities, it solved 50.3% of the total tasks when maximum reasoning levels were applied.
This is more than double the public evaluation score of 21.5% for GPT-5.6 Terra.
However, OpenAI explained that GPT-6.1 Sol and GPT-6 Astra remain more suitable for complex coding tasks that require a high degree of expertise.
Alongside performance improvements, security capabilities have also been strengthened, with the model evaluated as reaching "Critical"—the highest risk level for cybersecurity capabilities—under OpenAI's AI risk assessment and response framework, the "Preparedness Framework."
GPT-6.1 Terra is available to users of ChatGPT Team, Plus, Pro, Enterprise, and Edu, and developers can use it under the model name "gpt-6.1-terra" via the OpenAI API.
Standard API pricing is set at $1 per 1 million input tokens, $0.05 for cached inputs reusing previously processed data, and $5 for output.
(Photo courtesy of OpenAI, Yonhap News)
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.