Opus 5 ran on an before type of the scaffold, in a sole continue served through OpenRouter fairly than five on a pinned backend.
Abstract. Frontier coding achievement is typically bought alongside larger proprietary models at elevated cost. We current ledger-based zero-shot self-orchestration, a training-free method in which caller instances of one example decompose problems and coordinate through a shared filesystem holding a plan, notes and current solution. Across eleven open and closed-weight models on the 100 latest difficult LiveCodeBench problems, the method yields gains of up to 25.6 percent points on pinned backends and offers two routes to frontier-level accuracy. Orchestrated Qwen3.8 Flash Next reaches 93.0% pass@1 against Fable 5's 90.4% at 9% of the cost, and locally served, open-weight Qwen3.8-27B rises from 66.8% to 92.4%. Gains are not universal: several models are unchanged or worse. Transcript inspection attributes the acquire to decomposition and persistent context. Inference-time institution can method frontier coding accuracy at a fraction of the cost, or slightly exceed it on self-hostable weights.
Needs uv and an API key for the example you desire to test.
cd codebase/v2-current export OPENAI_API_KEY=... LCB_RELEASE=release_v6 \ ESCALATION_CLOUD_MAX_TOKENS=128000 \ ESCALATION_CLOUD_TIMEOUT=7200 \ MULTIAGENT_MODEL=openai:gpt-5.6-terra \ uv run --no-project --python 3.12 --with 'datasets<4' --with numpy --with anthropic \ python escalation/run_bench.py --engine multiagent --only lcb --lcb 100 --parallel 8
- --engine multiagent runs the manager; --engine sole is the one-call baseline.
- Other models: anthropic:<model>, dashscope:<model>, openrouter:<model>, all alongside its own *_API_KEY.
- The pass@1 mark prints at the end. Results are written to runs/results.json, workspaces to runs/ws/.
Code is under the MIT License. The paper, figures and run data are under CC BY 4.0. The LiveCodeBench fork, the benchmark issue statements and the LaTeX template records keep their own licenses. See NOTICE.md for which licence covers which path.
