claude-model-optimizerlisted

# Claude Model Optimizer Route work across Claude Opus 5, Claude Fable 5, and Codex/GPT-5.6 Sol without turning every task into a multi-model ceremony. ## Default Policy Start complex Claude Code work with **Opus 5 at `high` effort**. It is the everyday default for planning, production implementation, debugging, code review, knowledge work, and agentic tool use. Escalate to **Fable 5** when the task needs Anthropic's highest available capability: unusually ambiguous architecture, the hardest long-horizon work, deep or obscure knowledge, or a judgment call where Opus has exposed a real capability gap. Use **Codex/GPT-5.6 Sol** as a frontier peer when an independent perspective, a different tool harness, large-context evidence gathering, persistent execution, or browser/computer-use automation is useful. It is not limited to throwaway scripts, and its output is evidence rather than authority. Prefer one capable model end to end when that is enough. Add another model only when it has a distinct job. For large work that splits into independent packets, Fable 5 can coordinate bounded workers while retaining planning and integration judgment. Use this as a measured exception to the one-owner default, not as a standing swarm. ## Model Roles | Model | Default role | Reach for it when | Watch for | |---|---|---|---| | Opus 5 | Everyday Claude Code lead | Complex coding, planning, refactors, enterprise work, visual tasks, review, and multi-step execution | Over-verification, s