Qwen4 with Various Sizes?
The Qwen team has shown the new model architecture. It’s great. But a big question now is:
Will Qwen4 Release Models in a Variety of Sizes?
I feel that Qwen no longer prioritizes small models in the range from 1B to 36B like Qwen3.5. They seem to be targeting the Top 1 LLM in the world. But this requires a very large model, with trillions of parameters, given the current LLM architecture.
I know they need “money” to train models, and they need “money” from API services to cover the costs. This is normal, not bad.
However, I don’t think everyone is ready to build and maintain a small “machine” to run a small model, e.g., 8B, 16B, or 32B, to serve a small number of users through their AI agents.
In short, the big part of the cake belongs to big companies, and the small part is left for small startups with very limited resources.
Currently, Qwen and Google are two companies that release high-quality small models for the community. But
Will they stop following this strategy?
You want smaller sizes then 35BA3? Why? That runs basically on any PC already.
Are you trying to run LLMs on a phone? If so, isn't it smarter to just use your main PC as a server and access the model from the phone?