Local 27B model replaced his coding API: a Qwen workflow and cost breakdown
Training-Respect8066 · reddit · 2026-09-24
A Redditor details running a local Qwen coding model (27B) instead of a paid API:
- Usage pattern: the model thinks a lot, so the trick is to let it run unsupervised; it can complete complex refactors with good decisions on its own.
- Setup: Q4KS quant with Q80 context quantization; Swift-Qwen is faster but loops more often. He uses the Pi agent without MCP, keeping only bash plus read/write/edit tools — the edit tool is the weak link, causing indentation-related retries. Docker serves as sandbox, and the agent runs on a Raspberry Pi.
- Cost: about 2.4 cents per 1M input tokens and 70 cents per 1M output tokens on his hardware, comparable to the cheapest providers on nano-gpt.com.
More from coding & agent
- Why inference speed will be AI products' biggest competitive edge in 1–2 years — itsOmSarraf_ · 2026-11-11
- What happened when I moved vector retrieval into Snowflake SQL instead of orchestrating RAG outside — dima806_dima · 2026-11-11
- OpenAI engineer runs Codex for 26 days straight hunting Atlantis, spawning "vibe archaeology" — Polymarket · 2026-11-11
- He replaced a $5,000/month ghostwriter with 11 Claude Skills and earned $47,000 from one post — aryanXmahajan · 2026-11-11
- Refero Styles ships 2,000+ DESIGN.md files to give Claude Code real design taste — bbssppllvv · 2026-11-11
- Agent circuit breaker: openclaw shorts bot halts itself after 3 failed experiments — Rich_Standard_1724 · 2026-11-11