MERA Boosts Small LLM Agent Capabilities and Efficiency

Yuhang Yao, Zeyu Wang, Wanyi Chen, Tongyun Yang, Yuhang Han, Jie Xiao, Chengke Bao, Tianyi Zhao, Lynn Ai, Eric Yang, Tianyu Shi· August 12, 2026 View original

Key takeaways

  • MERA improves small LLM capabilities for agentic tasks through iterative skill adaptation.
  • It uses replay of failed invocations and teacher demonstrations to distill recurring procedures.
  • Cost-calibrated routing with verifier-backed fallback ensures quality and reduces inference costs.
  • Empirical results show significant pass rate increases for small models and substantial cost savings.

Who benefits

Software DevelopmentAI/ML PlatformsCustomer ServiceRoboticsEnterprise AI

Summary

MERA is a new framework that enhances small language models (LLMs) for agentic systems by iteratively improving their "skills" through replay of failed invocations and distillation of procedures. It combines model evolution with cost-calibrated routing to achieve significant capability gains and cost savings.

A novel framework called MERA (Model Evolution and Routing with Skill Adaptation) has been introduced to significantly improve the capabilities and efficiency of small language models (LLMs) within agentic systems. Unlike prior routing methods that merely assign tasks to existing small or large models, MERA actively enhances the small model itself. It operates through a multi-cycle adaptation process where failed small model invocations are replayed, verified by a larger "teacher" model, and then distilled into an iteratively updated "SkillBook." This SkillBook is used to fine-tune a student LoRA adapter, effectively teaching the smaller model new skills. The improved student model is then deployed behind a cost-calibrated router, which includes a verifier-backed fallback to the larger model, ensuring task quality is maintained. Empirical results demonstrate substantial gains: a Qwen2.5-Coder-1.5B model saw its pass rate on HumanEval+MBPP increase from 28.7% to 49.7% after four adaptation cycles, while deployed policies retained high pass rates at significantly reduced costs compared to always using a large model.

Why it matters

Professionals in AI engineering, product development, and operations should care about MERA as it offers a practical approach to deploy more capable and cost-effective LLM agents by actively improving smaller models rather than just routing around their limitations.

How to implement this in your domain

  1. 1Implement a multi-cycle adaptation loop for fine-tuning smaller LLMs based on feedback from larger models or human verification.
  2. 2Develop a "SkillBook" mechanism to distill recurring procedures and knowledge into a smaller model's training data.
  3. 3Design and deploy cost-calibrated routing policies for LLM agents, incorporating verifier-backed fallbacks for quality assurance.
  4. 4Explore LoRA adapter fine-tuning strategies for efficient skill transfer and model updates in agentic systems.

Original post by Yuhang Yao, Zeyu Wang, Wanyi Chen, Tongyun Yang, Yuhang Han, Jie Xiao, Chengke Bao, Tianyi Zhao, Lynn Ai, Eric Yang, Tianyu Shi

"arXiv:2608.10333v1 Announce Type: new Abstract: LLM agents execute heterogeneous sequences of model calls within a single task: some invocations require careful reasoning, while others are structured steps such as formatting or tool-argument construction. Prior routing methods ex…"

View on X

Originally posted by Yuhang Yao, Zeyu Wang, Wanyi Chen, Tongyun Yang, Yuhang Han, Jie Xiao, Chengke Bao, Tianyi Zhao, Lynn Ai, Eric Yang, Tianyu Shi on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses