Q: Why did Kagi start enforcing the fair use policy?
A: The policy was enforced due to excessive use. For instance, the top 10 users accounted for approximately 14% of the total costs, with some individuals consistently using up to 50 million tokens per week on the most advanced models. Our profit margins are already quite narrow. 95% of users should never hit any usage limits.
Don't think it's fair to call users problematic when they were using the product as advertised. "Unlimited" has a meaning.
I'm sympathetic to that argument, for sure, but it's also just a branding-label to not necessarily be taken literally. There must always be a limit to everything as there's only so much energy in the universe: so, the word 'unlimited', in every real-world physical context, is still with limits.
Read the T&Cs should always be the the advice.
Sure sure sure, but I'm not abusing the "Unlimited" service. I'm just asking AI questions every now and then. I'm a normal user doing normal usage and I have no idea if I will be hitting these limits or not.
There's a subtle difference between using something for 5 hours per day vs causing the heat death of the unvierse.
Metred use, with all parties being informed and honest with wording, is the fair and ethical solution. I absolutely abhor how companies are allowed to change meanings of words, then run behind 'muh conditions' when they lose out on their gamble.
Replace 'heat death of the universe' with 'the available funds of the organisation'. Nobody should be so naive to think that any 'unlimited' service is unlimited in the same way as there's an unlimited set of natural-numbers, or any other mathematically pure meaning. There are plenty of words where the precise meaning isn't used any more ('myriad' is one that jumps to mind right now, nobody uses it to mean 10,000 any more).
I'm sure if there was a word that meant: "effectively unlimited for the majority, but there's a limit for the extreme outliers" I'm sure it would be used as an alternative, I don't know of one?
I agree that the absolute upper-limits should be upfront. So the local definition of 'unlimited' is clear. I believe that's what Kagi is moving toward, if I'm reading FAQ correctly.
I hate to break it to you, but I think the cat is well and truly out of the bag on that one!
You can't say you get 10 apples for a dollar and only give 9. You can say best, ultimate Apples because they are not quantifiable.
If we make a quick back of the envelope estimation: in the outlier case if those 3 million tokens are mostly output tokens and you always used an advanced model like GPT 4.1 which costs $8 per 1 million output tokens you would be close to hitting the limit that the plan provides ($24 out of $25 worth of tokens).
In pretty much most other scenario's (including a higher proportion of input vs output tokens and mixing in cheaper models) you could be a long way from hitting the limit. For example if you used half of those tokens on GPT 4.1 Mini instead of GPT 4.1 you'd only be roughly halfway to your limit ($14 out of 25$ worth of tokens).
As it stands, I use up to 2M tokens per month but have no clue how much this amount of tokens (across various models) costs.
And 5% hitting the limit and not having a way to pay for usage past the limit (yet) is kind of scary. Especially as I feel like I use AI more than my peers.
I agree about the need for appropriate wording and advertising, but other than that, the new limits seem entirely reasonable and in line with what other aggregators like Abacus and Poe are doing. The paid plans of the major AI labs themselves always have usage limits too. It simply can't work any other way if you include costly models in the mix.