Qwen 3.6 27B agent gets much smarter after KV cache quantization change
Jordanthecomeback · reddit · 2026-07-28
A Reddit user reports a large quality jump after changing a local Qwen 3.6 27B agent’s quantization settings, especially the KV cache.
They were troubleshooting a slow Apple-silicon setup and moved from the standard iq4NL quant to a q4M heretic build, while also following Claude’s advice to use Q8 KV cache. The agent became noticeably better at tool use, recall, and following the system prompt, with roughly the same runtime but a similar memory footprint. The poster suspects that less aggressive quantization at either the model or KV level would help even more.
More from coding & agent
- 10 open-source repos that let your AI agent reverse-engineer apps and games — victor_explore · 2026-10-16
- The new CTO job: agent management and QA loops, 'cable management' instead of coding — HankYeomans · 2026-10-16
- Why company AI agents go stale: adoption needs in-house knowledge, not just better agents — EdenEmarco177 · 2026-10-16
- Open-source MCP lets coding agents search Obsidian's 9,000-plugin directory before reinventing wheels — pitfa19 · 2026-10-16
- Cheng Lou open-sources Handwired: hand-written weights turn code into training-free neural nets — dotey · 2026-10-16
- Building a portable context vault in Notion so every AI agent knows you from day one — ivanhzhao · 2026-10-16