An AI agent that reads the web pays two distinct kinds of cost, according to a new analysis. The first is the retrieval charge: what a search or fetch API bills per request. The second is model processing: the content passed from retrieved pages into an AI model consumes tokens, and reasoning plus generated responses add further token usage.
TinyFish, a search service, is now offering free search to agents. That eliminates the retrieval side of the equation, but it does not make the agent cheap to run. The model still has to read every page it fetches, and every token it processes carries a price.
The distinction matters for developers choosing tools. A free search API can lower the barrier to entry, but the dominant cost for most agents will likely remain the model's token consumption. The source notes this trade-off explicitly, without claiming that TinyFish solves the larger expense.