AI / favorite chat LLMs

Favorite LLMs

My favorite LLM's which are below 15GB:
(the QAT quantizations seem to work the best/fastest on my machine)

• Gemma 4 E2B / E4B
• Gemma 4 26B A4B
• GLM 4.6V Flash (about half the speed of Gemma 4 26B A4B)
• Bonsai 8B
• Ornith 1.5 35B A3B (about half the speed of Gemma 4 26B A4B)
• Qwen 3.8 27B

Some of my favorite LLM's above 15 GB:

• Seed OSS 36B Instruct (Q3_K_M, GGUF, 16.41GB)
  Good for creating a roleplaying game with multiple players. Not the fastest though, needs to think for some time, but one of the few "small" LLM/AI's that can actually keep track of multiple persons/players.
• Qwen 3.8 27B

Smallest LLMS

These are extremely small but still somewhat coherent:

• LFM2 2.6B
• LFM2.5
• Baguettotron (by PleIAs)
  (321 million parameters generalist Small Reasoning Model based on llama)
  GOOD - Tiny, about 350 MB (8 bit), 185 MB (4 bit), extremely fast, smart analysis/thinking about your question
  BAD - Tiny 4K context,  cannot act upon it's own (smart) analyses/planning, cannot create stories or roleplay, quickly gets confused
• SmolLM3 3b (thinking model) (8 bit MLX, 3.28 GB) / (4 bit MLX, 1.75GB)