Model handles proposal drafting, blueprint reading and more

Standard API price cut 50% vs. predecessors; available via ChatGPT Work and Codex

OpenAI's GPT-6.1 Terra scored higher than its predecessor GPT-5.6 Terra on Agents' Last Exam v1, a benchmark measuring professional task performance, while lowering the cost per task. [Provided by OpenAI]
OpenAI's GPT-6.1 Terra scored higher than its predecessor GPT-5.6 Terra on Agents' Last Exam v1, a benchmark measuring professional task performance, while lowering the cost per task. [Provided by OpenAI]

OpenAI has unveiled GPT-6.1 Terra, an AI model designed to deliver stronger task performance at a lower cost. The model extends support to accounting workflows while cutting standard application programming interface (API) prices in half compared with existing models, targeting enterprise demand for automation.

OpenAI announced Tuesday that GPT-6.1 Terra enhances both everyday communication and computer-based task capabilities across a broad range of workplace functions.

The company said the model is built to handle high volumes of work efficiently. Supported capabilities include drafting proposals, interpreting technical blueprints, filling out forms, editing video, performing visual reasoning, executing computer tasks and writing. Developers can use these features to automate large-scale workflows — for example, reading the contents of an invoice and entering the details into accounting software.

Pricing comes in at roughly half that of predecessor models. The standard API price is 50 percent lower than both GPT-5.6 Terra and GPT-6.1 Sol. The model is available through ChatGPT Work, Codex and the OpenAI API.

Task capacity has also expanded. OpenAI said users can process more tasks within the same usage limits on ChatGPT Work and Codex compared with GPT-5.6 Terra. The increase in throughput is even greater relative to GPT-6.1 Sol and GPT-6 Astra.

Computer task performance has also improved. On Agents' Last Exam, GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra at a medium reasoning level. The benchmark includes professional tasks such as converting architectural drawings into 3D models and reviewing animation frame by frame to produce timecode logs. The cost came in at about one-third that of GPT-5.6 Terra. Compared with GPT-6 Astra, GPT-6.1 Terra delivered similar performance at one-tenth the price per token.

On MMMU-Pro, a visual reasoning benchmark, GPT-6.1 Terra outperformed GPT-5.6 Terra across all reasoning settings. On OSWorld 2.1, which evaluates multi-step task performance using desktop applications, it surpassed GPT-5.6 Terra's top score at a medium reasoning level — also at a lower cost per task.

OpenAI's GPT-6.1 Terra solved 50.3 percent of tasks on Terminal-Bench 4.0, a coding performance benchmark, at maximum reasoning — more than double the 21.5 percent recorded by its predecessor GPT-5.6 Terra. [Provided by OpenAI]
OpenAI's GPT-6.1 Terra solved 50.3 percent of tasks on Terminal-Bench 4.0, a coding performance benchmark, at maximum reasoning — more than double the 21.5 percent recorded by its predecessor GPT-5.6 Terra. [Provided by OpenAI]

OpenAI also invested heavily in reducing hallucinations and factual errors. According to the company, in an internal evaluation using questions designed to elicit mistakes, GPT-6.1 Terra recorded an error rate of 16.7 percent at a low reasoning setting — about 37 percentage points below GPT-5.6 Terra's 53.6 percent. At maximum reasoning, the error rate fell to 4.3 percent, 18 percentage points lower than its predecessor and approaching the levels of GPT-6 Astra and GPT-6.1 Sol.

Coding performance has also been strengthened. On Terminal-Bench 4.0, which measures a coding agent's problem-solving ability in a command-line environment, GPT-6.1 Terra solved 50.3 percent of tasks at maximum reasoning — more than double GPT-5.6 Terra's published score of 21.5 percent. However, GPT-6.1 Sol and GPT-6 Astra remain better suited for the most complex coding tasks.

Work on the model's safety systems is ongoing. GPT-6.1 Terra has reached the "Critical" threshold for cybersecurity capability under OpenAI's Preparedness Framework. OpenAI said it is applying the same safety measures used in GPT-6 Astra to GPT-6.1 Terra.

In addition, OpenAI's evaluations found that GPT-6.1 Terra adhered more closely to the scope of user-requested tasks than GPT-5.6 Terra did. It also communicated more clearly to users about the model's limitations and any factors blocking task progress. The model showed strong resistance in prompt injection tests involving malicious instructions designed to steer it away from user tasks.

GPT-6.1 Terra is available to Plus, Pro, Enterprise and Edu users of ChatGPT Work and Codex. Developers can access it via the OpenAI API under the model name "gpt-6.1-terra." Standard API pricing is set at $1 per 1 million input tokens, $0.05 per 1 million cached input tokens and $5 per 1 million output tokens.


chami@heraldcorp.com