Piccolo: Co-Evolving Decision Models and Data through Recursive Self-Improvement
Abstract
We present Piccolo, a family of multimodal decision models (0.8B to 27B) that coevolves model capabilities and training data via Agentic Data Engineering and recursive self-improvement (RSI). The models predict probabilities over candidate answers for general-purpose text and visual judgments, agent harnesses, data workflows, and interactive games. Piccolo-27B tops the Decision Index Vision Full leaderboard with a score of 75.59, while Piccolo-9B and Piccolo-4B achieve 68.35 and 67.46, respectively. All three rank first on the Private benchmark in their respective model-size categories. Research agents analyze model behavior, construct training data, conduct experiments, and use the results to refine subsequent proposals. Semantic slices and a shared research memory connect these actions across the model family's development. With training recipes comprising hundreds of thousands of examples, data selection and feedback-driven refinement are central to the self-improvement process. Our large-scale RSI exploration yields three counter-intuitive insights for decision-model development: (1) Backbone-recipe co-adaptation outweighs raw model recency: earlier backbones can surpass newer generations on visual tasks under identical recipes; (2) Targeted data curation enables 4B-9B models to rival 27B baselines on specialized workflows; and (3) Continuous loss reduction can decouple from and even degrade discrete decision accuracy. Together with the released weights and tools from the Data-Juicer and AgentScope ecosystems, Piccolo provides an actionable blueprint for building frontier decision models and scaling data-driven RSI.
Community
🚀 Introducing Piccolo: #1 Multimodal Decision Model Series (0.8B–27B) Co-Evolved by AI Research Agents!
Autonomous systems bottleneck on fast, reliable decisions—routing workflows, approving tool calls, filtering data, and interacting in real time. Piccolo solves this with a fast System-1 decision interface (Choice, Score, Yes/No) and a ~99ms median latency.
✨ Highlights & Key Takeaways:
- 🏆 SOTA Across Scales: Piccolo-27B tops Decision Index Vision (75.59 Full / 70.97 Private), while Piccolo-9B (65.04) and Piccolo-4B (63.49) rank #1 in their size classes on Private.
- 🔄 Agentic Data Engineering (RSI): Built via an autonomous research loop (
Data-Juicer+AgentScope) across ~600 experimental runs and a 663K semantically annotated data pool. - đź’ˇ Actionable Post-Training Insights: We detail why continuous loss reduction can degrade decision accuracy, why earlier backbones can beat newer ones under fixed recipes, and how recurrent layer reuse unlocks massive gains on agent harnesses.
- 🎮 Versatile Applications: From enterprise guardrails and data pipelines to real-time action selection in Black Myth: Wukong, Zenless Zone Zero, Mario, and 2048.
🎥 Real-Time Decision Demos (Black Myth: Wukong & ZZZ):
đź”— Try Piccolo Today:
- 🎮 Live Playground: https://decisionplayground.agentscope.io/
- 🤗 HF Models (0.8B to 27B): https://huggingface.co/collections/agentscope-ai/piccolo
We're excited to hear your feedback—drop a comment below if you have any questions!
Get this paper in your agent:
hf papers read 2611.24651 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper

