OpenAI introduces GPT-6.1 Terra, an AI model that integrates visual reasoning, desktop automation, and software engineering capabilities while reducing operational costs This article explores gpt terra ai. . The release comes with a heightened security designation: OpenAI classifies Terra as having a Critical cybersecurity capability under its Preparedness Framework, applying safeguards inherited from GPT-6 Astra and GPT-6.1 Sol.
The model targets high-volume professional workflows, including interpreting technical drawings, processing invoices, completing forms, and editing videos.
OpenAI announces that standard API prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol, positioning the release around affordable automation rather than maximum frontier performance. On Agents' Last Exam, which evaluates professional computer-based tasks, Terra scores 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort while costing approximately one-third as much. Terminal-Bench 4.0 demonstrates an impressive 50.3% success rate on command-line engineering tasks, outperforming GPT-5.6 Terra's 21.5% score.
Terra achieved a 97.10% ACE-normalized mean capability score on ExploitBench, which evaluates the exploitation of known vulnerabilities using descriptions, source code, and patches without reference exploits. On a separate internal evaluation based on recently disclosed vulnerabilities, Terra encountered arbitrary code execution in 24.5% of cases, compared to 21.5% for GPT-6.1 Sol and 31.5% for GPT-6 Astra. Explore for your team.
