Achieves a 19x reduction in inference cost and 16x in latency for agentic workflows by evolving hybrid LLM-and-code pipelines.
March 23, 2026
Original Paper
HyEvo: Self-Evolving Hybrid Agentic Workflows for Efficient Reasoning
arXiv · 2603.19639
The Takeaway
Instead of relying on homogeneous LLM-only reasoning, this framework automatically synthesizes workflows that offload predictable operations to deterministic code nodes. This provides a massive leap in efficiency for practitioners deploying complex reasoning agents where cost and speed are current bottlenecks.
From the abstract
Although agentic workflows have demonstrated strong potential for solving complex tasks, existing automated generation methods remain inefficient and underperform, as they rely on predefined operator libraries and homogeneous LLM-only workflows in which all task-level computation is performed through probabilistic inference. To address these limitations, we propose HyEvo, an automated workflow-generation framework that leverages heterogeneous atomic synthesis. HyEvo integrates probabilistic LLM