https://x.com/EMostaque/status/1729609312601887109?s=20
$1 => $100k+ flat for all models we release depending on company size/type, then upsell support, early access, custom models, other services.
I think that aligns with growth of market as a base to cover model costs and is a strong recurring revenue stream while being predictable.
This incentivises us to release more good models of every type to drive more subscriptions, much like Netflix with hit TV shows. So we are doing a dozen languages, common models improved as well as state of the art ones.
We also have commercial variants eg our www.stableaudio.com model that per licensing we can't release open.
Usually companies have an open core business model for source code, but models are a bit different: https://en.wikipedia.org/wiki/Open-core_model
It will be interesting to see how others do it, particularly with incredibly strong models from Chinese companies coming through.
To add to this, it is a valid strategy to give away something for free so attention - and therefore power - is taken away from a competitor. A quick way for everyone to forget OpenAI exists is for another nation or business conglomerate to release a model that is "good enough". This slows growth for businesses trying to profit from it.
I'd bet on one of four things
1. Early access for a fee. Get the latest model at a cost, they're later fully open sourced.
2. Sell commercial use licenses. Get and validate the model works for you easily, then buy a license. Split by company revenue if you want to
3. Sell hosted versions - these things require annoying tech to host, paying for someone else to manage running machines can help (possibly combined with a license around not setting up a resale of the LLM)
4. Sell the service of producing (and possibly hosting) custom models. Prove your worth with the open source ones, so you're a trusted person to come to in order to get a model custom made.
1. Offering a paid hosted solution down the road
2. Offering bigger / better models that cost money to use commercially
3. Consulting services around their models
4. Licensing their models to companies that want to host them as part of a paid offering
IMO the whole advantage of small, open source models is to avoid paying for hosted solutions and additional privacy / security, so it is a bit at odds with commercialization.