Dynamo: Dynamic Skill-Tool Evolution for Vision-Language Agents
For practitioners of vision-language agents, Dynamo offers a lightweight, training-free alternative to RL-based adaptation that achieves competitive gains without weight updates.
Dynamo is a training-free framework that adapts frozen vision-language models by evolving reusable reasoning skills and visual tools from a small labeled subset, improving accuracy by an average of +5.6 across 20 model-benchmark settings and closing 65-99% of the RL gap with minimal compute.
Improving vision-language models (VLMs) on visual reasoning typically requires retraining or hand-designed prompts and tools. We present Dynamo, a training-free framework that adapts a frozen VLM without any weight updates. On a small labeled training subset, the agent inspects its own correct and incorrect attempts and evolves two complementary capabilities: reusable reasoning skills for cognitive bottlenecks, and executable visual tools for perceptual ones. Each generated tool is paired with a skill that specifies when to invoke it, and both capability types accumulate in a persistent library. Across four visual reasoning benchmarks and five VLM backbones, Dynamo improves direct inference on all 20 model--benchmark settings (avg. +5.6 acc). When the tool set is given in advance, the framework learns when to call each tool, and per-step tool choice improves on every tested backbone. Against task-specific RL (VTool-R1, DeepEyes), Dynamo closes 65--99% of the RL gap at a fraction of the compute, and combines additively with RL when available.