Get Started
Home
Topics
Search
Library
Research questionWhen should a multi-step LLM agent escalate from a cheaper to a larger model?A cheaper model can reduce inference cost but may drift into persistent failure, while a larger model is more expensive. The routing challenge is recognizing when continued use of the cheaper model is unlikely to succeed.
AI
AI Agents
Evaluation & Benchmarks
Inference Optimization
Small / On-device Models
Latest papersRecent research connected to this question, newest first.TACIT-Switch: Cost-Aware Model Escalation for LLM Agents from Censored SupervisionThe source learns handoff policies from annotated intervention trajectories and evaluates them in a controlled multi-step simulation and on ALFWorld and DABench. The deployment policy requires no teacher, but the evidence is limited to these routing settings and tasks.research paper · Sep 4, 2026
Related questions
How can AI agents adapt execution routes as runtime evidence invalidates their planned continuation?How can LLM agents stay safe during multi-step execution when both policy and runtime harness shape behavior?How can we reliably locate and classify failures in long LLM-agent trajectories?How can long-horizon LLM agents learn when to group actions without overcommitting?