HASTE: A Hierarchical Multi-Agent System for Transfer-Efficient ML Engineering Across Competitions
By
[Submitted on 29 Jun 2026 (v1), last revised 1 Jul 2026 (this version, v2)]
Summary
This paper presents HASTE (Hierarchical Accumulation of Skills for Transfer-Efficient ML Engineering), a multi-agent system that organizes cross-competition knowledge into three scope tiers (global, domain, and competition-specific) to avoid cold-start inefficiencies in ML engineering agents. The system uses an orchestrator to coordinate domain specialists and promote learning between tiers via LLM-driven abstraction. Results show tiered loading achieves a 100% medal rate vs 62.5% for flat loading across 8 competitions, and warm starts use 52% fewer refinement iterations. On the full MLE-Bench Lite benchmark (22 competitions), HASTE reaches a 77.3% medal rate using Claude Sonnet 4.6.
Source
Twitter / XHASTE: A Hierarchical Multi-Agent System for Transfer-Efficient ML Engineering Across Competitionsarxiv.orgKey quotes
· 4 pulledML engineering agents waste compute rediscovering known techniques because every competition is a cold start.
Tiered loading achieves a 100% medal rate while flat loading reaches only 62.5%, the same medal rate as loading no skills, and consumes 2x the output tokens.
These results suggest that better knowledge organization can partly substitute for model strength and compute budget in ML-engineering agents.
Warm starts use 52% fewer refinement iterations, and the fraction of proposed changes kept by the agent rises from 42% at low inventory to 85% once 50+ skills are available.
You might also wanna read
SkillsBench: A Benchmark for Evaluating AI Agent Skills Across Diverse Tasks
SkillsBench is a new benchmark for evaluating how well AI agent skills work across diverse tasks. The benchmark includes 86 tasks across 11
arxiv.org·4mo ago
MAKER System Solves Million-Step LLM Task with Zero Errors Through Extreme Decomposition
Researchers have developed MAKER, the first system to successfully solve a task requiring over one million LLM steps with zero errors, addre
arxiv.org·7mo ago
Arbor: A Multi-Agent Framework Using Tree Search for Autonomous LLM Inference Optimization
Arbor is a multi-agent framework that uses structured tree search as a cognition layer for autonomous agents operating in large, stateful ac
arxiv.org·21d ago
Qwen-AgentWorld: Language World Models for Simulating Agentic Environments Across 7 Domains
This paper introduces Qwen-AgentWorld, a family of language world models (35B-A3B and 397B-A17B) designed to simulate agentic environments a
arxiv.org·12d ago
A Field Guide to Production-Ready AI Agents: Context Windows, Security, and Drift Monitoring
Karl Mehta presents a field guide for building production-ready AI agents, focusing on four key engineering challenges: context-window disci
OpenClaw-Skill: A Collective Skill Tree Search Framework for LLM-Based AI Agents
A new research paper introduces OpenClaw-Skill, a Collective Skill Tree Search (CSTS) framework that addresses key limitations in LLM-based
clawbeat.co·15d ago
Comments
Sign in to join the conversation.
No comments yet. Be the first.