hybrid-llm-routing-for-23-stage-pipelines

For AutoResearchClaw’s 23-stage pipeline, use Opus 4.6 only for the 8 critical reasoning stages (hypothesis generation, experiment design, paper writing, peer review simulation — stages 2, 7, 8, 9, 14, 15, 17, 18) and Haiku 4.5 for the 15 routine stages (topic init, resource planning, export, status checks). AutoResearchClaw natively supports primary + fallback model configuration. Stay within the Anthropic ecosystem (Haiku not GPT-4o-mini) to avoid managing external API keys and billing.