How much does it cost to run an always-on agent?

Nothing extra by default, from OpenAI: its own docs say “conversations with your dot don’t count toward your ChatGPT usage limits” at all. Only once it hands a task to Work or Codex does that task start drawing from those products’ normal usage limits, the same as if you’d used them yourself.

Google doesn’t split it out either. A Gemini agent run draws from the same pooled subscription quota as the rest of Gemini Enterprise, under the project-level spend limit that also covers the Gemini Enterprise app and its coding tools. Past that quota, usage stops unless you’ve turned on pay-as-you-go overage, which bills whatever you use with no pooled limit at all.

So the honest answer is that neither vendor prices the agent on its own. What you pay for is the model and tool calls it makes while it runs, the same spend you’d watch for any agent that keeps working with nobody in the loop to notice a cost creeping up.