Remove AI Character Limit at least for Ultimate Plan

mackid1993

The Ultimate plan is priced on par with ChatGPT plus, I feel like at least on this plan the artificial character limit should be removed. I was trying to parse a large XML file for virt manager and ChatGPT handled it just fine but it was a few thousand characters too large for Kagi. When I run into roadblocks like this it makes me want to go back to ChatGPT although I'd rather support Kagi's mission.

Especially when programming or working with a very large text file and needing to ask questions against it the character limit as high as it is can still be restrictive. At the pricing of the Ultimate plan it shouldn't be more limited than competing services.

Vlad

mackid1993 There are other differences worth noting.

ChatGPT+ and other competing services also have usage limits (certain number of messages per hour), while we do not (Kagi Assistant is basically currently unlimited).

ChatGPT+ allows you access to just OpenAI models, while Kagi Assistant allows access to all state of the art models (Claude 3 Opus, Gemini, ...).

Finally processing a message at for example Gemini token limit (1M) would cost $7 per message.

Open to suggestion, just thought complete context should be given first.

mackid1993

Vlad Perhaps the limit could be dependent on the model, obviously $7 a message isn't sustainable so capping Gemini would make sense, but for models that aren't as costly increasing the cap and making this limit clear and removing the cap for less expensive models to open up Assistant for use cases where a lot of context is needed would be super helpful while hopefully not putting a big dent in revenue if this is limited to just ultimate members as well as determined per the cost per model.

Vlad

mackid1993 Processing a full context message for GPT-4 (32k tokens) is ~$2 USD, and this is without output tokens which are twice as expensive. 10 messages like this and you are over what you paid for Ultimate plan, which also includes full access to Kagi search.

mackid1993

Vlad Fair enough. I didn't realize how costly it was. Thanks for the information. Hopefully as competition increases costs will come down. Also eventually Nvidia will have some competition and the cost of the hardware will come down as well.

gladiator2339

Have to disagree, only because I like Kagi so much. It will never be feasible. I'm already concerned about Ultimate Plan's future as is. It's pays for itself easily and I hope it continues to do so.

You can ask to remember this: and then continue on. Context window is crazy with newer models.

OldMan973471

Would this be resolved by allowing us to provide our own AI API keys for our favorite LLM? There are times I would like to work with massive data sets but I don't want to pay the $20/mo - would love to use Kagi's interface with my own OpenAI/Claude/______ API key and pay for what I use.