Laguna S 2.1 has been made to be specifically capable as a coding model. It's another entry in the class of models such as Deepseek V4 Flash and Hy3, which runs on a single DGX Spark or Strix Halo machine, with plenty of VRAM to spare for KV Cache. At Q4_K_M quant it's far smaller than either Hy3 at Q3 or V4 Flash at Q2 , and it comes in plain old GGUF format, so it doesn't need any sort of special framework to run (such as DS4 for Deepseek). Initial tests with it show really fantastic coding performance - I'll be putting it through the paces over the next week. This Youtuber provides some initial examples:

https://www.youtube.com/watch?v=ERCdFgTXNY4

It's also great to a US company providing some competitive self-hostable models.