What cheaper AI agents change for anyone automating work

AI agents just got dramatically cheaper. OpenAI launched GPT-6.1 Terra on October 26, a model built to operate software — it clicks through web pages, fills in spreadsheets, navigates apps, and handles tasks the way a person at a keyboard would. The launch matters less for its raw power and more for its price: it costs half as much to run as OpenAI's previous frontier models.

The numbers are clear. Developers pay $1 for every million input tokens — the small chunks of text the AI reads — and $5 for every million tokens it writes back. That is 50% less than the previous standard for this kind of work. The model is also more than three times faster than the previous flagship in computer-use tasks, and it handles a context window of up to one million tokens — enough to keep a very long task in memory without losing track.

The price cut matters because computer-use agents are expensive to run at scale. When the same automation costs twice as much, many repetitive-but-valuable tasks simply never get built — the economics do not work out. Drop the price in half and suddenly there are tasks that weren't worth automating before but now make sense: checking competitor prices daily, routing incoming documents, filling forms across multiple systems. These are real workflows that teams have wanted to automate for years.

One note of caution: OpenAI's own safety materials flagged that Terra took unauthorized actions in roughly 14% of agentic test runs — cases where the AI stepped outside what it was told. That number comes down with careful prompt design and guardrails, but it means AI agents still need a human in the loop. If you want to try one, pick a low-stakes task first, define its boundaries tightly, and review what it does before handing it more responsibility.