@tpop123 Of course it's possible. I just prefer products that are flat rates (e.g. unlimited phone data plans). I don't like always thinking about my usage.
The downside of flat-rate for both the user and business is that the pricing is based on average usage and customer LTV, meaning that customers who use the product far less are subsidizing the ones who use it the most.
@summersunshine1217 Risk is the biggest issue. What happens if you leak keys? You could suddenly go out of business drowning in lawsuits.
But yes companies are already offering exactly this type of model. You could even allow customers to pick their model to better tune cost and context window.
@summersunshine1217 Definitely leaking keys. You should have your subscriptions give a fixed amount of access which can be implicit or explicit. As in “personal use only” or “1000 requests per month”. Either way you need to have usage and rate limits. If you are letting people rawdog OpenAI on your dime, something as simple as frontend bug sending too many requests could knock you out.
@summersunshine1217 Sales is difficult as now your customer need 2 contracts, which it also has to manage. Easier if you resell it as part of the package. And even easier if you price it in your own product
@summersunshine1217 We're doing this in B2B our users enter their openai api key. The only alternative to that is if we charged them instead for it, so they'd have to deposit upfront the amount which would eliminate liability.
@summersunshine1217 You're missing that everyone struggles with capacity planning.
Which is why even when you have usage based pricing there always needs to a whole range of grace periods to allow people to still benefit from your product when they exceed the limit.
There's a reason it never took off except for specific situations e.g. logging, metrics.
@summersunshine1217 Downside of byo keys is that you now make your sale more complicated. You’ll need the customer to sign up with OpenAI or azure for the key + with you. Two sets of contracts etc. just a headache for the customer and for you. I see the byo key only for large enterprises with long sales cycles and they really really care about owning it all, at what point they probably want your software “on prem”
@summersunshine1217 I have thought about this. But, this is something openAI should support. Like let me log into AI apps using my own openAI account, like you login through Google.
Then I would pay a monthly subscription to OpenAi or ise any other OpenAi pricing model and log into any gpt based apps I want.
@summersunshine1217 well at this point one could also have a low-tier option that lets users use their own cloud services api keys.
but this isn’t the point. if you’re using an ai model to run your app, it’s because it’s a tool you’re leveraging today to fulfill the solution you’re providing.
because your app should most likely be solving a problem, which means that the tools used don’t really matter. tools change, all the time.
unless of course your app is literally another chat with ai.
@summersunshine1217 So many issues.
- rate limits & performance
- security / compliance
- most businesses use azure or GCP. Having customers put in their openAI key won’t help
- multi-model. Many tools use multiple models to improve performance. Not very thing is just openai
@lillian2014 Say I have an app that provides a nicer interface for copywriting with ChatGPT. The user can sign up for $30/month as a normal subscription plan, or they can elect for $5/month but they must enter their OpenAI API key in the settings to use the app.
@summersunshine1217 Well isn’t that two different products then? Because former is a whole app essentially, and the latter is basically a UI. Do I understand correctly?
@lillian2014 I have a very similar product to OP. It's not technically two apps, as the only (yet entirely fundamental) difference is what api key I send to openai/ openrouter. Everything else is identical. Users see the same thing, get the same quality of results, and have their data stored on my backend the same. It's just a matter of the user choosing to pay for your service and llm through you, or your service and pay llm directly.
