AI
AgentBaton: Efficient Model Collaboration in Agentic Tasks
Junlong Li, Jindong Jiang, Yonggan Fu, et al. · Submitted 25 Oct 2026 · 9 upvotes yesterday
Preprint · not peer reviewedTrending
The authors test handing a long AI agent task from one model to another mid-way. Starting with an expensive model and switching to a cheap one at the right moment worked well; the reverse did not. They built a small detector that learns when to switch.
Key finding: Across 11 model pairs, API costs fell by 65.9% on average while accuracy largely held, with gains of up to 8.8 points.
Why it matters: A simple, one-time switch could make running AI agents much cheaper.