OpenCode coding test: Qwen 3.6 vs Gemma 4

Qwen 3.6 and Gemma 4 receive the same task: build a cross-platform Markdown viewer with editing capabilities using Tauri. The speaker compares the 27B Qwen model with the 31B Gemma model, both dense, through OpenCode. They run sequentially on a desktop accessed from a MacBook over the local network.

This local LLM comparison tests a personal workflow rather than an objective benchmark. Each model gets the same project description, writes an implementation plan, then attempts the entire project in one pass. Qwen takes about four minutes to plan and 46 minutes to implement; Gemma takes about two and a half minutes to plan and 20 minutes to implement.

Both applications need manual launch fixes. Qwen omits server startup configuration and uses an older Tauri method. Its app then supports editing and live preview, but formatting buttons do not respond. Gemma needs filesystem plugins added to its configuration. Its app opens Markdown files and switches between editing and preview modes, but omits formatting toolbar buttons despite including them in its plan.

The speaker prefers Qwen's more detailed plan and notes Gemma's shorter completion time. Neither becomes the sole coding assistant: the stated choice is to keep using both.