i have been using Gemma4 for general purpose, and Qwen3-Coder-Next for coding

i’m always interested in small models (under ~8b) too for CPU or phone use

i may start using non local models like GLM5.2 or Kimi, since it would be nice to have “claude-level” LLM output (i don’t use proprietary models)

Gemma4 is designed to fit on phones, could try that

vivaldi is proprietary, bonus evil points for being chromium-based