Qwen3.6 Tested: Quantization Matters
_TheWolfOfWalmart_ · reddit · 2026-07-11
A Reddit user shared their test of Qwen3.6 35B-A3B: using a single prompt in opencode to generate a "beautiful, relaxing flight simulator," outputting a single HTML file with mountains, clouds, and infinite procedural terrain.
They first had the model plan in plan mode, then asked it to implement the plan without alterations. Initially unimpressed, switching the inference config from Q4KM on GPU to Q80 on CPU made the model feel "far beyond expectations"—slower, but worth it.
More from Infra
- Spomin: live KV cache compaction squeezes 500k tokens of context into 180k resident — wgaca2 · 2026-09-11
- PiPNN nearest-neighbor search wins three awards, up to 78x faster index building — khademinori · 2026-09-11
- M.2-Oculink eGPU Link Silently Downgrades to PCIe Gen1 — Here's How to Check — El_90 · 2026-09-11
- DeepSeek launches V4.1-Flash with 1M-token context and 4x smaller KV-cache — matlabulous · 2026-09-11
- What Can You Still Run on 8GB VRAM? User Asks for Small Models With Tool Use — riceinmybelly · 2026-09-11
- Spain's hourly 80% renewable matching rules clash as France fast-tracks 700MW sites, UK cuts grid queues — eherrerosj · 2026-09-11