OpenAI's new intermediate model replaces GPT-5.6 Terra: billed $1 per million tokens in entry, it divides its error rate on agent tasks by three.
OpenAI launched October 26th GPT-6.1 Terra, successor to GPT-5.6 Terra in its range. It is located between Luna, the most economical option, and the high-end models Sol and Astra. Its specialty: the use of the computer, i.e. reading what appears on the screen and then acting in the applications, browser or desktop. OpenAI cites the filling of forms, video editing, the reading of technical drawings and the automation of repetitive tasks.
Net differences with previous generation
On the Benchmark Agents' Last Exam, the model earns 15 points on its predecessor with average reasoning. Its error rate falls from 53.6% to 16.7% in low reasoning and 4.3% at maximum effort. In programming, it gets 50.3% on Terminal-Bench 4.0, more than double the 21.5% of GPT-5.6 Terra. OpenAI also states that it completes its tasks about three times faster than GPT-6.1 Sol and GPT-6 Astra.
The price drops by half compared to GPT-5.6 Terra and Sol: $1 per million tokens on entry and $5 on exit, i.e. one tenth of Astra's fare. The template is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, as well as via the API under the name gpt-6.1-terra. For the most complex code yards, OpenAI continues to recommend Sol or Astra.
Read the full article on the original source
Go to the source →