Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness

Researchers introduced a training-free method called ledger-based zero-shot self-orchestration that enables multiple instances of a single AI model to collaborate on coding problems through shared filesystem coordination. On hard coding benchmarks, the approach achieved significant improvements, with Qwen3.8-27B models reaching 92.4% accuracy (up from 66.8%) and matching or exceeding Claude Fable 5's performance at a fraction of the cost. The method demonstrates that inference-time orchestration can achieve frontier-level coding accuracy without requiring larger proprietary models or expensive training procedures.