Qwen3.6 35B A3B vs Gemma4 26B A4B: MoE Vibe-Coding Face-Off
Two MoE models take the same vibe-coding challenge: Qwen3.6 35B A3B vs Gemma4 26B A4B on Unsloth Q6_K_XL + llama.cpp across 4 prompts.
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?
Comments
Weekly digest
The best of vibe coding, AI agents and open source — once a week, no spam.
Related articles
All articlesAlibaba Qwen3.6-35B-A3B: Sparse MoE Model with 3B Active Params
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.
Local Browser AI Agent in a Chrome Extension with Gemma4 & WebGPU
Nico Martin built a fully local AI agent that runs in the browser using Gemma4, WebGPU, and Transformers.js to search history, summarize pages, and manage tabs.
Qwythos-9B-Claude-Mythos-5 Released with 1M Token Context Window
Empero unveils Qwythos-9B, a Qwen3.5-9B-based reasoning model fine-tuned on 500M+ tokens of Claude Mythos/Fable data and supporting up to 1M context.