Reproduce “AgentRouter: Heterogeneous Model Routing for Cost-Optimal Multi-Step Agentic Workflows” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rudrendu Kumar Paul, Sourav Nandy
AgentRouter routes steps of a multi-step AI workflow to one of four model tiers, using task and context signals and trajectory context to account for dependencies. Its classifier selects the cheapest tier meeting a quality threshold, with frontier-model fallback. On held-out benchmarks, it reduced cost 71.8% while retaining 97.3% of frontier-only quality; results require step sequences to be known or estimated.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.