QuickerSite Blog

By the time I click the save button below, this post is outdated. Things are rapidly changing in 2026.

I did a bit of research.

Below you can find 13 SVG files that were all generated with one and the same prompt by 13 different AI LLMs:

I used 3 local LLMs: Gemma 4 e4b, Qwen 2.5 Coder 7b and Qwen 3.5 9b. Let's be honest. They are useless.

But... take a look at the Qwen 3.6 Plus variant. This is a proprietary model hosted by Alibaba, the well known Chinese web shop.

However, Qwen 3.6 27B was recently (April 2026) released as an open source and free model that can be installed locally, on powerful consumer hardware. It appears to be a gamechanger for local agentic coding. I can't wait to give this LLM a try.

Meantime I developed my own AI chatbot on https://github.com/PieterCooreman/Synapse. Synapse connects to locally installed LLMs (via Ollama or LM Studio) and tries to get the best results possible. 

I'll be back with more news about Qwen 3.6 27B. I have high expectations. I recently bought a 128GB RAM desktop horse. It should run this LLM at a reasonable speed.

 Meantime, check out this document, generated by my locally installed Gemma 4 b31 (model by Google). Gemma 4 is pretty good at authoring texts. But it's not very good at coding. Qwen is much better with code. Both models can be installed simultaneously on Ollama or LM Studio, giving you more flexibility when deploying a local AI Lab.