Local Qwen as junior coder vs agent: a $20 Codex + 24GB VRAM hybrid workflow
mesolimbic-bliss · reddit · 2026-11-13
The poster runs a hybrid workflow: a $20 Codex subscription plus a 24GB VRAM local rig. Codex's Luna delegates most coding to a local "junior dev" Qwen 3.8-27B, reviews and fixes its output, then Terra audits the finished project. Now they're considering flipping it — handing the full multi-phase spec straight to Qwen and having Luna just audit — and ask which approach (local model as initial coder vs. as called agent) wins on quality vs. usage limits.
More from coding & agent
- New research tackles the agent experience problem: knowledge trapped in weights — khademinori · 2026-11-13
- YC F26's Acyclic Labs open-sources infra to run 10,000-agent swarms in one session — ycombinator · 2026-11-13
- Floci: free local AWS emulator with 31.6k GitHub stars for faster integration tests — JeremyCMorgan · 2026-11-13
- Open-source REM tool reverse-engineers competitors' ads, SEO and social playbook from one URL — moeinteractive · 2026-11-13
- Codex agent shrinks a 27M-param transformer to 11.46M with equal loss via weight sharing — yoshiyama_akira · 2026-11-13
- Google open-sources Harness²: Agent = Model + Harness, recursive agent harnessing framework — 2C_ornot2C · 2026-11-13