OpenAI Releases GPT-6.1 Terra: Faster Computer Use at Half the API Price

What happened

Last month's GPT-6 refresh covered Sol and Luna, leaving the mid-tier Terra out; the October 26 release fills that gap. OpenAI positions GPT-6.1 Terra as an everyday-work model: it reads what is on screen, acts across apps and writes natural text, aimed at workday tasks such as drafting proposals, reading technical drawings, filling out forms and editing videos.

It is available immediately: ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and for developers via the API as gpt-6.1-terra. OpenAI says in its announcement that Terra substantially outperforms GPT-5.6 Terra on Agents' Last Exam, a test of how agents use office software, matches GPT-6 Astra at one-tenth the price per token, and finishes tasks over 3x faster on average than both Astra and GPT-6.1 Sol; all of these are OpenAI-reported figures.

Price and accuracy move together: standard API prices are $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens, half the GPT-5.6 Terra rates. On OpenAI's internal error-inducing prompt test, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7%, and at maximum effort fell a further 18 percentage points to 4.3%; on Terminal-Bench 4.0 the maximum-effort score is 50.3%, again OpenAI-reported.

Key facts

1. Release and access: Released October 26, 2026; available now in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and via the API as gpt-6.1-terra.
2. Positioning: The GPT-6 family's mid-tier model: last month's refresh covered Sol and Luna, and Terra now fills the gap; the family runs Luna, Terra, Sol and Astra from lowest to highest price and capability.
3. Pricing: Standard API prices of $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens, half the GPT-5.6 Terra rates.
4. Reported performance: Substantially above GPT-5.6 Terra on Agents' Last Exam; OpenAI says tasks run over 3x faster on average than GPT-6 Astra and GPT-6.1 Sol at one-tenth of Astra's price per token; Terminal-Bench 4.0 maximum-effort score of 50.3%.
5. Reported accuracy: On the internal error-inducing prompt test, the error rate fell from 53.6% to 16.7% at low reasoning effort and to 4.3% at maximum effort.