OpenAI launches GPT-6.1 Terra, an intermediate model presented as fast and token-efficient, designed for high-volume tasks. Announced in a press release published on Monday 26 October 2026, it succeeds GPT-5.6 Terra and stands out mainly for its speed and reduced cost in the API.
A “fast and efficient” model, designed for automation
OpenAI describes GPT-6.1 Terra as “fast and efficient” and as a model “that combines detailed visual reasoning and precise computer usage with clearer, more natural writing”. Like GPT-6.1 Sol, unveiled at the DevDay 2026 conference, it is first deployed in ChatGPT Work and Codex. According to the American company, it performs tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. The release qualifies, however: “For the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the best choice”.
On Agents’ Last Exam, a benchmark composed of professional tasks performed on a computer, GPT-6.1 Terra clearly outpaces its predecessor. It even rivals GPT-6 Astra “with a medium reasoning effort”, for a much lower cost. When its reasoning effort is set to maximum, OpenAI indicates that it consumes fewer tokens (the text units processed by the model) than Claude Opus 5.5, one of Anthropic’s flagship models.
Fewer factual errors
OpenAI also highlights the reduction in error count. A hallucination rate, i.e., false answers presented as true, is mentioned in the release. “On our internal benchmark of queries designed to induce errors, the error rate with low reasoning effort dropped from 53,6 % with GPT-5.6 Terra to 16,7 % with GPT-6.1 Terra,” the company specifies. These are results reported by OpenAI on its own test, which have not been independently verified in the source.
The model also reaches the “critical” threshold in cybersecurity, i.e., the highest level of OpenAI’s internal risk‑assessment grid. It therefore inherits the safeguards designed for GPT‑6 Astra. Regarding alignment, the company states that it deviates less often from the scope of the task assigned to it and that it “more honestly signals its limits and any potential roadblocks to the user”.
Reduced pricing in the API
Since 26 October 2026, Plus, Pro, Enterprise and Edu subscribers can use GPT-6.1 Terra in ChatGPT Work and Codex. It has not yet been deployed in ChatGPT. Developers can access it via OpenAI’s API, under the identifier gpt-6.1-terra. The announced price is $1 per million input tokens and $5 per million output tokens. “Its API pricing is 50 % lower than that of GPT‑5.6 Terra and GPT‑6.1 Sol,” the California‑based company notes.
