OpenAI halves the price of its mid-range model and points it at agent-driven office work
The new GPT-6.1 Terra is meant to read screens better and operate software more precisely, at half the price of its predecessor.
OpenAI on Monday released GPT-6.1 Terra, the middle model in its range between the cheap Luna and the more powerful Sol and Astra. It replaces GPT-5.6 Terra and costs half as much, according to 9to5Mac. The model is aimed at everyday tasks such as drafting quotes, interpreting technical drawings and filling in forms.
Through the API, Terra costs 1 dollar per million input tokens and 5 dollars per million output tokens, InfoWorld reports. Heise cites independent measurements by Artificial Analysis: the estimated cost per task at maximum reasoning effort drops from 1.40 to 0.40 dollars, and the model far less often answers without knowing the answer. OpenAI itself reports a sharp fall in the error rate on an internal test.
The model is rolling out gradually in ChatGPT Work and Codex for paying subscribers, but not in the regular chat. Analysts quoted by InfoWorld warn that cheaper compute only reduces part of the cost of AI agents: reviewing and integrating their work still takes time.