AgentFloor: How Far Up the tool use Ladder Can Small Open-Weight Models Go?

2026-04-30

AI Summary

This study addresses the capability boundaries between large and small language models in agent systems by introducing AgentFloor, a deterministic benchmark comprising 30 tasks organized into a six-level capability hierarchy. The authors systematically evaluate 16 open-source models (ranging from 0.27B to 32B parameters) alongside GPT-5 across dimensions including instruction following, tool usage, multi-step coordination, and long-horizon planning. Through structured benchmarking, cross-model comparisons, and intervention analyses based on 16,542 human evaluations, the work establishes the first clear delineation of applicability boundaries for model sizes within agent workflows and proposes a complexity-aware hybrid scheduling principle. Results show that the strongest open-source model matches GPT-5 in overall performance while offering lower cost and faster inference, though frontier models retain an advantage in long-horizon tasks requiring sustained constraint tracking.

Abstract

Production agentic systems make many model calls per user request, and most of those calls are short, structured, and routine. This raises a practical routing question that existing evaluations do not directly answer: which parts of an agent workflow truly require large frontier intelligence, and which can be handled by smaller models? We introduce AgentFloor, a deterministic 30-task benchmark organized as a six-tier capability ladder, spanning instruction following, tool use, multi-step coordination, and long-horizon planning under persistent constraints. We evaluate 16 open-weight models, from 0.27B to 32B parameters, alongside GPT-5 across 16,542 scored runs. Our results reveal a clear boundary of model necessity. Small and mid-sized open-weight models are already sufficient for much of the short-horizon, structured tool use work that dominates real agent pipelines, and in aggregate, the strongest open-weight model matches GPT-5 on our benchmark while being substantially cheaper and faster to run. The gap appears most clearly on long-horizon planning tasks that require sustained coordination and reliable constraint tracking over many steps, where frontier models still hold an advantage, though neither side reaches strong reliability. We also find that this boundary is not explained by scale alone: some failures respond to targeted interventions, but the effects are model-specific rather than universal. These findings suggest a practical design principle for agentic systems: use smaller open-weight models for the broad base of routine actions, and reserve large frontier models for the narrower class of tasks that truly demand deeper planning and control. We release the benchmark, harness, sweep configurations, and full run corpus.

Problem

• agent workflow
• model routing
• tool use
• long-horizon planning
• open-weight models

Innovation

• AgentFloor
• tool use
• model routing
• long-horizon planning
• open-weight models