760 600 queries per day. That’s about 8.8 queries per second.
Per user it’s about 15 queries per day. I’m sure there will be some that are incredibly active and some that aren’t active at all, but 15 per day seems quite reasonable.
I think so too. Apparently I have an overall mean of 26 queries/day, with the lowest month the past year being 19 queries/day and the highest 35 queries/day. Most of this is during the weekday with software development work, but I also make web searches for all kinds of other things in my spare time.
I genuinely thought it would be higher, but I suppose bang patterns don't count.
(Posting this mostly so that people who are curious about subscribing to Kagi can get a sense of how many queries they're likely to need to use.)
Date (UTC) AI Tokens AI Cost (USD) Searches
Jun 2025 0 0.000 141
May 2025 0 0.000 743
Apr 2025 0 0.000 723
Mar 2025 0 0.000 621
Feb 2025 0 0.000 556
Jan 2025 10,692 0.000 1,189
Dec 2024 0 0.000 805
If you append ? to the end of a query, kagi runs the query through an llm that cites its sources. It very rarely hallucinates at this point, but I still click through to check.
I keep trying chatgpt style products, but rarely reach for them because I don’t really have a concrete use case for them. I use ? all the time, code completion at work and regularly generate images. Generic chat can do those things but is never as good as specialized tools I have access to.
My guess is that purpose-built specialized expert AIs are going to end up being more useful than the “everything” products like the chatgpt UI, but time will tell, I guess.
Is the AI search part of the normal search plan or charged extra? Because atm, I'm still using Perplexity on a trial plan, but would like to try Kagi if it did the same, but used a better underlying search engine.
I’ve been measuring my token count and I believe they are making a small amount on me even though I’m not holding back my AI usage. I do pick the cheaper and faster LLMs when I can, but mostly to avoid the waiting of heavy models with extended thinking.
You're probably aware, but they recently added the actual cost of used tokens on the Billing page. This is much better than the token count on its own.
By "I've been measuring my token count", I do mean "I've been reading my billing page and comparing my token count month over month", but thanks for pointing that out so it's clear.
What the billing page doesn't go into detail about is number of searches caused by The Assistant, number of FastGPT searches, and number of regular searches. I'm curious because I'd like to track my tendencies; whether I am slowly using AI more than search, or if it stays the same. And FastGPT is a grayzone.
Per user it’s about 15 queries per day. I’m sure there will be some that are incredibly active and some that aren’t active at all, but 15 per day seems quite reasonable.