Are small quantized models worth running as coding sub-agents? Users say they loop and fail tool calls
Eastern_Doughnut_222 · reddit · 2026-11-18
A Reddit user asks whether it's worth running small local quantized models as sub-agents alongside a main coding model.
Key takeaways:
- Small quants (2-3bit range) in the author's experience tend to loop, fail tool-calls, or get stuck
- The exception: Q8 quantized QFN has been "pretty astounding" when set up properly
- The author has a second machine capable of running 2-3bit quants of 27B models and is weighing whether the volatility is worth it
- Prior sub-agent experience was via API on full-sized enterprise models, a very different cost profile
- Leaning conclusion: small-quant sub-agents are risky unless the quant quality is high or the use case is narrow and specialized
More from coding & agent
- Microsoft's Autopilot Agent Is Built on OpenClaw, Now in Preview for Frontier Customers — pswider · 2026-11-18
- SimplePost Gives ChatGPT and Claude a Publishing Button for 10 Social Networks — haltakov · 2026-11-18
- One set of Obsidian notes feeds Claude Code, Codex and Gemini alike — dSebastien · 2026-11-18
- Obsidian Starter Kit gives Claude Code, Codex and Gemini the same persistent context — dSebastien · 2026-11-18
- AI tools start from zero at every prompt — the missing piece is persistent context — dSebastien · 2026-11-18
- 27 vibe-coded Finder clones appeared in a month, most already dead — KaKi_87 · 2026-11-18