The paper introduces a model‑adaptive definition of tool necessity that accounts for each LLM’s empirical performance, contrasting it with observed tool‑call behavior across four models on arithmetic and factual QA datasets. It finds substantial mismatches—26.5‑54.0% for arithmetic and 30.8‑41.8% for factual QA—between when tools are needed and when they are actually used. By probing hidden states, the authors show that the main source of mismatch lies in the transition from internal cognition (recognizing tool necessity) to action (making a tool call), revealing a knowing‑doing gap in LLM tool use.
arXiv:2609.14157v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed with external tools that extend what they can do beyond their own knowledge. Tools help on tasks...
The paper investigates how reinforcement learning can cause large language model agents to adopt shortcut policies for tool use, relying on superficial prompt cues rather than actual task needs. By creating synthetic environments that mix factual QA and math reasoning, the authors show that agents often invoke tools when cues are present, even when those tools are unnecessary, with spurious invocation rates rising up to 39%. They find that shortcut learning occurs mainly when agents have already mastered the target tool and that semantic alignment between cues and tools amplifies the effect. To counter this, they propose a dense, decision-level reward where an LLM judge assesses tool necessity, which reduces cue-driven tool use while maintaining performance.
arXiv:2603.18614v2 Announce Type: replace Abstract: Tool-augmented large language models (LLMs) must tightly couple multi-step reasoning with external actions, yet existing benchmarks often confound...
arXiv:2610.09624v1 Announce Type: cross Abstract: Tool calling, invoking external tools on demand, is central to agentic LLMs, yet the mechanism that decides whether a model calls a tool or responds...
arXiv:2607. 04686v1 Announce Type: cross Abstract: Tool calling is central to modern language model agents, but aggregate benchmark scores often hide where tool use fails.
Self-evolving agents can continually improve their behavior, while tools define the executable action space through which they interact with the environment. However, exposing the full tool library to...
arXiv:2610.02330v1 Announce Type: new Abstract: Large language models (LLMs) rely on long-horizon tool invocation sequences for complex tasks, where each invocation can alter the task state and condi...
The paper investigates automated failure attribution (AFA) in large language model (LLM)-based agentic systems, where multiple agents collaborate on complex tasks. It finds that LLMs contain sufficient internal knowledge to attribute failures but often fail to output the correct step, revealing a knows‑but‑won’t‑say gap. To address this, the authors introduce grafting—a manual intervention that leverages latent model knowledge—showing it consistently outperforms direct prompting across several LLMs and benchmarks.
The paper investigates how the representation of tool interfaces—referred to as a tool schema—affects the performance of large language model agents. By systematically transforming a native tool schema with nine operators, the authors demonstrate that agents exhibit significant variability in success rates across different but functionally equivalent schema variants, a phenomenon they call schema bias. Experiments on eleven LLMs, including closed models, reveal that success can range from complete failure to 97% solely due to schema differences, and that training only mitigates bias when the specific variant is included in the training data.
The paper introduces SMITH, a reinforcement learning framework that jointly trains large language models to create and use tools within a single policy. By alternating between build and use tasks and employing separate reward signals for schema, code, and outcome failures, SMITH enables a 4B Qwen3 model to achieve state‑of‑the‑art accuracy on procedural reasoning benchmarks, outperforming larger untrained models and improving performance on downstream tasks when its tools are applied.
ToolRobustBench is a stage-wise diagnostic benchmark designed to evaluate and diagnose failures in tool‑calling agents, which are large language models that select tools, provide structured arguments, and interpret tool feedback. The benchmark aligns four perturbation families—tool‑interface, user‑intent, tool‑output/observation, and runtime‑environment—with the tool‑use pipeline, attributing failures to specific stages such as tool selection, schema grounding, argument binding, and feedback handling. Experiments across 15,456 instances, 7 models, and 16 local tools reveal that while overall performance is high, robustness degrades significantly, especially under tool‑output/observation perturbations, and mixed‑family perturbations produce non‑additive failure patterns.
arXiv:2606. 15508v1 Announce Type: new Abstract: Tool-augmented large language model agents increasingly operate over large tool libraries, but existing evaluations often focus on whether a model can call a tool correctly rather than how the visible tool menu shapes reliability, efficiency, and safety-relevant risk exposure.