It would be nice to have better documentation regarding LLMs in Kagi. Benchmark is very nice, but unless someone's experienced with the topic it doesn't offer much.
Like, how does tokens in messages work withing current soft limit of 250k: e.g
To estimate the number of messages, let's break this down:
Average message length varies but typically ranges from 100-300 tokens
Let's use 200 tokens as a reasonable average per message
With a 250,000 token weekly limit
Basic calculation: 250,000 ÷ 200 = 1,250 messages
So you could send approximately 1,250 messages per week, or about 178 messages per day.
Keep in mind that:
Longer, more complex messages use more tokens
System prompts and AI responses also count toward the limit
The actual number could be lower if your messages tend to be longer
How does the context work. What to do if a model refuses to answer, e.g. Claude is notorious for refusals, and Mistral Large is very "open-minded".
It would be also nice to have the docs visible and accessible from the Assistant page, so beginners to LLMs can learn how to use them better, resulting in less frustration for them and saving costs for Kagi, since users will get results they wish faster.
