I saw a bunch of decent projects using Qwen 2507-4B, and it seems like another branch of the LLM progress tree
where instead of building one massive 400B model, we use a suite of 1B–30B models, each specially designed for specific tasks.
This approach seems more feasible, cost-effective, and something that can be iterated on in a very short amount of time.

There probably 5-7 new games that do look like that but you choose to ignore them so you can compare it something entirely different
Btw re7 looks Fucking great

>>715699474
Palworld has better pokemon than the best of the last 5 generations combined