reply
Thank you, I had no idea ollama was so shady! I will start using llama.cpp directly.
replyMore like `ollama launch claude --model qwen3.6:latest`
replyAlso you need to check your context size, Ollama default to 4K if <24 Gb of VRAM and you need 64K minimum if you want claude to be able to at least lift a finger.
If you're on a Mac, use the MLX backend versions which are considerably faster than the GGML based versions (including llama.cpp) and you don't need to fiddle with the context size. The models are `qwen3.6:35b-a3b-nvfp4`, `qwen3.6:35b-a3b-mxfp8`, and `qwen3.6:35b-a3b-mlx-bf16`.
replyI only have 16GB VRAM, and my system uses ~4GB from that. What are my options? I got this one: `Qwen3.6-35B-A3B-UD-IQ2_XXS.gguf`
replyhave you found a model that does this with usable speeds on an M2/M3?
replyOn a M4 MBP ollama's qwen3.5:35b-a3b-coding-nvfp4 runs incredibly fast when in the claude/codex harness. M2/M3 should be similar.
replyIt's incomparably faster than any other model (i.e. it's actually usable without cope). Caching makes a huge difference.