$ cat wiki/models/gpt-6-1-terra.md
GPT-6.1 Terra
Spec
| Attribute | Value |
|---|---|
| Developer | OpenAI |
| Released | 2026-10-26 |
| Announced | 2026-10-26 |
| Context window | unknown |
| Pricing | $1/M input · $0.05/M cached input · $5/M output |
| License | unknown |
| Availability | OpenAI API (gpt-6.1-terra); ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users |
| The announcement supplies standard API rates and access, but not context capacity or license terms. Dates follow the publisher's feed metadata. (announcement) (date) |
Release Date
OpenAI's feed dates the announcement to 2026-10-26; the article says access begins that day. (source) (availability)
Benchmarks
OpenAI reports 50.3% on Terminal-Bench 4.0 at maximum reasoning effort, versus GPT-5.6 Terra's published 21.5%. On its internal factuality evaluation, low-effort error rates fall from 53.6% to 16.7%; maximum-effort Terra reaches 4.3%. These are vendor measurements, not independently reproduced results. (source)
On Agents' Last Exam at medium effort, OpenAI reports a 15 percentage point gain over GPT-5.6 Terra at about one-third the cost. The captured prose does not provide absolute scores or complete harness settings; a relative improvement is not a task-success rate. See Eval Harness Configuration. (source)
Use Cases
OpenAI positions Terra for visual reasoning, desktop actions and writing, including invoice entry and interpreting drawings. The announcement also classifies it at the Critical cybersecurity threshold under Preparedness Framework, and claims improved scope adherence and prompt-injection resistance. The full linked system card was not reviewed here, so this page does not independently assess those safeguards. (source)
Compared To
Standard API rates are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol, according to OpenAI. Its article still recommends Sol and Astra for the most complex coding work. Token pricing and cost per completed task are separate comparisons. (source)
Sources
- Announcement prose: captured through the web fetcher; charts and the system card are outside this capture.
- Publisher feed metadata: publication date and canonical article URL.