Five Qwen3.8-27B models outperform Fable 5 Using GVS5H Harness
Summary
GVS5H benchmarks Five Qwen3.8-27B models against Claude Fable 5 on LiveCodeBench, delivering gains in accuracy and cost efficiency. The work highlights ledger-based zero-shot self-orchestration for coordinating multiple model instances and emphasizes open-weight models, with code and papers licensed under MIT and CC-BY-4.0.