Dev ships full racing game coded 100% by local Qwen 3.8 Flash Next after 1,911 prompts
hiImMate · reddit · 2026-11-14
A developer built a complete free-to-play TypeScript + three.js racing game entirely with a locally-run Qwen 3.8 Flash Next (Unsloth UDQ4XL quant) on a Strix Halo machine, using his own harness and a Gufo inference fork. No frontier model touched the code.
Project stats
- 1,911 prompts, 31,883 tool calls
- 20.4M tokens out, 3.07B tokens in (3.04B cached)
Verdict: If you can run it, he calls 3.8FN the best local model today — it resolved virtually every coding issue, though it's not built for 3D asset work.
Debugging tactics that worked
- Ask the model to paste the code for review: 70% of the time it spotted the bug itself
- Insist on measurement, validation and headless runs — read the numbers instead of reasoning from code
- Have it build keyboard-toggleable in-game logging tools, then hand it raw data for repro'd bugs
Anti-pattern: never hand it decisions. Treat it as a junior full of textbook knowledge that will take shortcuts whenever given too big a slice; scope must be explicit, and autonomous dev mode consistently produced lame results.
More from coding & agent
- Solo dev builds real-time conference translation platform for 70 languages with an agentic harness — evisoft · 2026-11-14
- Open-source Blender add-on runs NVIDIA Kimodo, ARDY and MotionBricks locally for AI motion generation — bdsqlsz · 2026-11-14
- TestPrune: pruning regression tests to unblock AI-driven autonomous bug fixing — burkov · 2026-11-14
- Context beats model choice: Databricks benchmark shows 91% vs 58% for data agents — CautiousUse8597 · 2026-11-14
- Developer skips localhost entirely: prompts go straight to production with AI agent safeguards — marclou · 2026-11-14
- HeyGen MCP demo: one ChatGPT prompt books tickets and sends an avatar birthday video — toolstelegraph · 2026-11-14