4-bit MLX quantization of bottlecapai/ThinkingCap-Qwen3.6-27B (Qwen3.6-27B finetuned for ~50% shorter thinking traces). See the original card for evals and recommended sampling params.

Quantized with mlx-vlm 0.6.4 (4-bit, group size 64 (mlx_vlm default), ~16 GB):

mlx_vlm convert \
  --hf-path bottlecapai/ThinkingCap-Qwen3.6-27B \
  --mlx-path ./ThinkingCap-Qwen3.6-27B-mlx-4bit \
  --quantize --q-bits 4

Runs in LM Studio (requires a recent MLX engine with qwen3_5_vision support).