1. Value of your service for customers is the gold key to business success. Private model can be an advantage but is usually not the key (unless your model is ChatGPT).
2. The most common failure reason is few people know and buy your product. There are so many open sourcing projects which nobody cares! Open-sourcing choice itself is not so important in startup early-stage, but will highly impact your business model when your startup grows.
3. Open sourcing can be a kind of advertising which help your product reach non-tech customers via their tech friends or KOL.

if you have a lot of data, you can try to train larger ones first, and solve the model size problem with all kinds of model compression or inference optimization methods.

For infamous papers, nobody cares it.

For famous papers, many people will reproduce it and make decisions based on its performance, which is a kind of human-based statistical testing.