OpenAI Releases GPT-6.1 Terra: For Office Tasks, With Computer Use
OpenAI has released GPT-6.1 Terra, positioned as a fast and efficient mid-tier model. It supports fine-grained visual reasoning, computer use, and more natural writing for office tasks such as drafting proposals, interpreting drawings, filling forms, and editing videos. The model is available to some users in ChatGPT Work and Codex, and developers can access it via API. Officially, it improves over GPT-5.6 Terra on professional tasks, visual reasoning, desktop operation, factuality, and coding benchmarks, with lower cost and higher efficiency.
OpenAI has released GPT-6.1 Terra, positioned as a fast and efficient mid-tier model that combines fine-grained visual reasoning, computer use, and more natural writing for office tasks such as drafting proposals, interpreting drawings, filling forms, and editing videos.
The model is available to ChatGPT Plus, Pro, Enterprise, and Edu users, with entry points including ChatGPT Work and Codex; developers can also call it via the OpenAI API using gpt-6.1-terra.
Officially, under medium reasoning effort on Agents' Last Exam, Terra scores 15% higher than GPT-5.6 Terra, and cost per task is about one-third; at highest reasoning effort, token-use efficiency is more than 3x that of Claude Opus 5.5, with unit price about one-quarter.
In visual and desktop operation tests, Terra exceeds GPT-5.6 Terra across all reasoning efforts on MMMU-Pro; at medium reasoning effort on OSWorld 2.1, it surpasses its best score. Internal error-induction tests show error rates drop to 16.7% at low reasoning effort and 4.3% at highest reasoning effort.
Coding tests show Terra solves 50.3% of tasks at highest reasoning effort on Terminal-Bench 4.0, higher than GPT-5.6 Terra; on Terminal-Bench-Science 0.1, it exceeds GPT-6.1 Sol.
On safety, Terra reaches Critical level in OpenAI Preparedness Framework cybersecurity assessment and carries over GPT-6 Astra safeguards. Officially, it is better at staying within task scope, explaining limitations, and resisting prompt injection.
KEY HIGHLIGHTS
Key highlights
GPT-6.1 Terra is positioned as a mid-tier model, supporting visual reasoning, computer use, and everyday office tasks.
ChatGPT Work, Codex, and API are open; developers can call it with gpt-6.1-terra.
Officially, professional-task scores improve, and cost is about one-third of GPT-5.6 Terra.
Safety assessment reaches Critical level and carries over GPT-6 Astra safeguards.