New research weโre excited about: MERA powered by
Evolve small models with SkillBook + routing + adapters so they take on more of the agent workload over time, instead of always defaulting to the biggest (and most expensive) model.
Faster agents. Lower costs. Better systems.
Paper + code ๐
๐ MERA: Model Evolution and Routing with Skill Adaptation for Agentic Systems at Scale
How can we make small models stronger for on-device agent deployment?
MERA uses stronger models to guide an iterative loop of RL/GRPO, skill learning, and router optimization. Student failures become verified demonstrations, reusable SkillBook procedures, and LoRA updates, helping the small model take on more work over time.
๐ฅ Results:
Qwen2.5-Coder-1.5B: 28.7% โ 49.7% coding pass
Qwen3.5-2B on TAU-2: 14/35 โ 18/35
Fine-tuned 2B matches an unadapted 4B model
Donโt just route around small models. Evolve them.
๐
๐ป