Open-weight models match Fable 5 on hard LiveCodeBench at a tenth of the cost: $5.76 vs $61.11
vecto_r · reddit · 2026-09-18
- A training-free manager-worker scaffold uses fresh instances of one model to decompose problems and coordinate, letting open-weight models approach or match Claude Fable 5 on hard LiveCodeBench.
- Gains from orchestration: Qwen3.8-FlashNext 84.2% → 93.0%; Qwen3.8-27B 69.2% → 92.4%; GPT-5.6-Terra 80.8% → 88.0%; GPT-5.6-Luna 70.4% → 81.2%; vs Fable 5's 90.4% single-call baseline.
- Cost gap is stark: orchestrated FlashNext runs $5.76 per pass against Fable 5's $61.11.
- Code and paper are public (GitHub GVS5H, arXiv 2608.26480).
More from coding & agent
- Free course covers the 4 hard problems of production generative media agents — Al_Grigor · 2026-11-04
- Quota reset night: Codex back to 100% while Claude shows just 10% left — rudrank · 2026-11-04
- Self-improving software is inevitable: closing the agent-observability loop — bibryam · 2026-11-04
- Building a 7-GPU 132GB VRAM local inference rig: consolidate on EPYC or split across two systems? — infectiousstupidity · 2026-11-04
- Founder of 50-person, 12-country team says he talks to AI agents more than humans — Scobleizer · 2026-11-04
- Pi launches a minimal agent harness; Zammoth adds AI engineer to Mac music app — solyarisoftware · 2026-11-04