TinyFish Removes Retrieval Fees for AI Agents Within Set Rate Limits
Web infrastructure firm TinyFish eliminates per-request search costs, though downstream model processing fees remain a factor for users.
Key points
- TinyFish offers free search and fetch requests within specific daily and hourly rate limits.
- The service uses a web-native AI model that runs on managed infrastructure without local GPUs.
- Downstream model inference costs persist because retrieved content consumes input tokens.
Web infrastructure company TinyFish has eliminated retrieval charges for its search and fetch APIs, provided users stay within defined rate limits. This change affects how AI agents access live web content without incurring per-request fees from the infrastructure provider.
According to MarkTechPost, the removal of retrieval fees does not eliminate all costs associated with web reading. Teams still pay for model processing when retrieved content enters an AI model’s context window. These input tokens, along with reasoning and generated responses, drive the remaining expenses for users.
Context and Background
TinyFish argues that organizations often underestimate how much retrieved content drives model costs while budgeting primarily for the model itself. The company’s platform uses a web-native AI model designed specifically for search, retrieval, and interaction with live websites. This model runs on TinyFish-managed infrastructure, utilizing a cloud browser fleet that renders pages and handles interactions without requiring local GPU resources.
Access to the platform is available via REST, Python and TypeScript SDKs, CLI, or MCP. In a performance snapshot from September 28, 2026, by Artificial Analysis, TinyFish Search (web) scored 71. The service cost $9.30 per 1,000 benchmark tasks, with an average time of 59.0 seconds per task, which includes both model and search time.
Among the 12 configurations displayed on the Artificial Analysis leaderboard, TinyFish had the lowest total cost. However, it also recorded the second-longest time per task. The highest score on that leaderboard was 80 for Perplexity Search (medium), while the lowest was TinyFish’s 71. A no-search baseline scored 33.
Who Is Affected
The free retrieval tier applies within specific rate limits: 30 requests per minute and 500 per hour for search, and 150 URLs per minute and 1,000 per day for fetch. Beyond these limits, standard billing likely applies, though the source material does not specify overage rates. The Agent product includes a built-in model that cannot be swapped, which is a noted trade-off for users seeking flexibility.
Enterprises are increasingly routing tasks to the cheapest capable models to lower price per token. This trend was highlighted when Stripe agreed to acquire OpenRouter, a gateway spanning over 400 models from more than 80 providers. While routing reduces token prices, it does not change the volume of tokens contained in a retrieved page, making page size a critical cost factor.
What should you do?
Organizations using AI agents for web retrieval should audit the size of returned pages to manage downstream token costs. Since retrieval is free within limits, the focus shifts to optimizing the content passed to the model. Users should monitor their usage against the stated rate limits to avoid unexpected charges.
What to do and how to stay safe: TinyFish
- Review the character count of fetched web pages to estimate input token consumption for your specific AI models.
- Monitor API usage closely to ensure you remain within the 500 hourly search and 1,000 daily fetch limits.
- Evaluate whether the fixed model in TinyFish’s Agent product meets your requirements, as it cannot be swapped.
- Compare total costs including inference against other providers, as free retrieval does not eliminate model processing fees.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Does TinyFish charge for search requests?
No, TinyFish offers $0 retrieval charges for search and fetch within its specified rate limits.
What costs remain after using TinyFish’s free search?
Users still incur inference costs for the downstream model, which bills for input tokens from retrieved content, reasoning, and output.
What are the rate limits for free access?
The limits are 30 requests per minute and 500 per hour for search, and 150 URLs per minute and 1,000 per day for fetch.
