Someone gave two MoE models the exact same vibe-coding challenge.
Qwen3.6 35B A3B (31.8 GB) vs Gemma4 26B A4B (23.3 GB)
Stack:
> Unsloth Q6_K_XL
> llama.cpp
> for each — sampling recommended in the model card
4 prompts, side by side. Who do you think will win?
Alibaba’s Qwen3.6-35B-A3B is a sparse MoE with 3B active params, strong coding and multimodal vision results, plus GGUF for local 4-bit runs.
New Qwen3.6 35B A3B distilled on Claude 4.6 Opus data. GGUF quantizations Q4_K_M to Q8_0—aimed at better logic and fewer loops.
Qwen3.6-27B is a 27B dense multimodal model that outperforms the 397B Qwen3.5 flagship on SWE-bench and Terminal-Bench, with Claude Code API setup.