TinyFish made its Search and Fetch APIs free within published limits, and the company’s reasoning is more interesting than the pricing change.
An agent reading the web pays twice. First comes the retrieval charge per request or page. Then the returned content enters the model context and gets billed as input tokens, with reasoning and output on top. TinyFish argues teams budget heavily for the first and underestimate the second.
The measurement that makes the argument
In a June evaluation across 45 URLs and five providers, TinyFish fetched a Daily Mail article with roughly 4,300 characters of body text. TinyFish returned 4,737 characters. Firecrawl returned 171,095, Exa 65,793 and Tavily 46,433.
At an illustrative $1 per 1M input tokens, that is about $0.0012 against $0.043 for the largest response, roughly 36 times more. At 42,774 tokens, a single Firecrawl page eats about a third of a 128K context window. Three such pages fill it.
Agent loops make it worse, because each turn resends the conversation. Caching discounts the rereading but does not free the window.
TinyFish scored 93 percent coverage on its own test, against 80 for Tavily, 73 for Exa, 62 for Firecrawl and 58 for Parallel. The obvious caveat is that TinyFish ran the test on itself across a small news-heavy sample.
On the September 28 Artificial Analysis leaderboard, TinyFish Search scored 71 at $9.30 per 1,000 tasks, the lowest total cost of 12 configurations but also the lowest score, with Perplexity Search medium at 80. Its run took about 59 seconds per task, near the slowest shown.
Limits are 30 searches per minute and 500 per hour, plus 150 fetches per minute and 1,000 per day, all still free at a zero wallet balance.
