[everforward]

I doubt that fits in the usage model for those pricing plans. They're priced assuming it's inconvenient and less used because of it. I keep hearing they're losing money anyways, so despite being triple or quadruple dipped you're still paying less than it costs to run the models.

[basch]

How would it cost Kagi anything, if I bring my own keys and all the costs are offloaded to my personal Google/ChatGPT account? I pay them the $10/month and get the Unlimited Pro access (but only to models I subsidize the cost of.) If anything it would save them money to offload my usage to my account vs using up some of their Flash tier quota.

It would be ideal for them for LLM access portability to let them offer higher end models and have the end consumer pay directly for the usage.

Id much rather pay for one AI License, and a small fee to each service I use, rather than paying a high tier AI Bonus price at every single service.

[macrocosmos]

I think they might be talking about the cost to the model providers.

[basch]

Then Google should offer Advanced Portable for $30 that includes being able to plug it into third party services. With some kind of policing to prevent it from being used as the entire backend to a service (for all users) itself.

[everforward]

Third party usage is too disparate for a single flat rate to make sense. Perplexity probably uses orders of magnitude more tokens than Kagi does (the former uses tokens on every search, the latter only does if you manually invoke the Assistant).

Then there's the disparity in how many third party services each user uses. Some will use one, some will use 50. Google could charge per integration, but that's basically the status quo.

[basch]

Why would the number of integrations be the price point and not token tiers? I might want to connect my account to 100 services to test them, but only perform a couple uses each before forgetting about it.

[everforward]

I agree that makes sense, but it makes potential subscribers to the third parties nervous. I like Kagi Assistant because my payment is capped.

I get nervous plugging my pay-as-you-go API keys into random software because of the risk they rack up a $1000 bill doing something I wouldn’t have paid $20 for them to do.

The other three things are that economies of scale make it cheaper for Kagi to buy a bajillion tokens, Google et al don’t want to manage the customer side of things (what service ate how many tokens?), and service providers don’t want you seeing their “magic” in your console. Seems like there’s a lot of power in the system instruction side of things, and Perplexity probably doesn’t want you seeing their prompt.

[yellowapple]

> I keep hearing they're losing money anyways

Hearing from whom?