TinyFish Waives Search API Fees for AI Agents Within Usage Caps

TinyFish has removed per-request billing for its AI agent search and fetch APIs, provided usage stays within established rate limits. According to MarkTechPost, this change means teams no longer pay a direct fee for each web page the agent retrieves, but they still pay for the data that enters their AI model.

In plain English

AI agents that browse the web incur two distinct types of expenses. The first is the retrieval charge, which is the fee paid to a service for fetching a webpage or running a search query. The second is model processing cost, which arises when the text from those fetched pages is fed into an artificial intelligence system.

This text consumes input tokens, while the agent’s reasoning and final answers consume additional output tokens. TinyFish argues that organizations often budget heavily for the model processing but underestimate how much the retrieved content drives those token costs. By removing the retrieval fee, TinyFish aims to highlight that the real expense lies in the volume of data processed by the model, not the act of fetching it.

The background

On September 28, 2026, the Artificial Analysis snapshot showed TinyFish Search (web) scoring 71 out of 100. This performance level cost $9.30 per 1,000 benchmark tasks, with an average completion time of 59.0 seconds per task. This score was lower than the highest displayed result, Perplexity Search (medium), which achieved an 80. The lowest displayed score was TinyFish Search (web) at 71, while a baseline configuration with no search capability scored 33. TinyFish’s solution runs on a web-native AI model designed for search and interaction with live sites. It uses a cloud browser fleet to render pages, requiring no local GPU.

What changes now

Teams using TinyFish can now make up to 30 search requests per minute and 500 per hour without a retrieval charge. For fetching specific URLs, the cap is 150 requests per minute and 1,000 per day. Within these limits, the retrieval cost is zero. However, the downstream model still incurs inference costs based on the size of the returned pages. TinyFish noted that page size matters significantly; for example, a Daily Mail article with roughly 4,300 characters in its body was tested against five providers on June 2, 2026. The company also noted that its Agent product uses a built-in model that cannot be swapped, unlike some competitors that allow model routing.

Key terms explained

Retrieval charges are fees paid to services for accessing web data, while model processing costs depend on how much text that data contains. Tokens are the units of text an AI model processes, and context windows define how much information the model can hold at once. Teams usually check whether their agents are fetching more data than necessary, as larger pages increase token counts. Monitoring the size of retrieved content helps predict processing costs, even when retrieval fees are waived. Checking default settings for data extraction tools can also reveal how much extra text, like ads or navigation menus, is being included in the payload.

What to do and how to stay safe: TinyFish

• Review your current AI agent workflows to identify how many web requests are made per minute and per hour.
• Estimate the average size of the web pages your agents retrieve to predict downstream token costs.
• Check whether your current provider allows swapping underlying models or if you are locked into a specific one.
• Monitor the performance scores of your chosen provider against industry benchmarks to ensure quality meets needs.

Step-by-step guide: Stop AI Chatbots Leaking Sensitive Business Data

General security guidance from the Vector Update newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Does TinyFish charge for search requests at all?

TinyFish charges zero for retrieval within its rate limits of 30 requests per minute and 500 per hour.

What is the performance score for TinyFish Search on the Artificial Analysis benchmark?

TinyFish Search (web) scored 71 on the September 28, 2026 Artificial Analysis snapshot.

Can I change the underlying AI model in the TinyFish Agent product?

No, the TinyFish Agent product uses a built-in model that cannot be swapped, according to the source material.