On Generalization in Agentic Tool Calling: CoreThink Agentic Reasoner and MAVEN Dataset

TL;DR

This paper evaluates large language models' ability to generalize in agentic tool-calling tasks, introduces a new challenging benchmark MAVEN, and proposes CoreThink, a symbolic reasoning framework that significantly improves performance without extra training.

Contribution

The paper introduces MAVEN, a new OOD benchmark for multi-step reasoning, and proposes CoreThink, a symbolic reasoning layer that enhances LLM generalization across diverse tool-use environments.

Findings

Most models score below 50% on MAVEN, indicating a large generalization gap.

CoreThink achieves 530% performance improvement over baselines.

CoreThink generalizes across all benchmarks without additional training.

Abstract

Generalization across Agentic tool-calling environments remains a key unsolved challenge in developing reliable agentic reasoning systems. While large language models (LLMs) demonstrate strong performance on isolated benchmarks, their ability to transfer reasoning strategies and co-ordinate tools across diverse domains is poorly understood. In this work, we conduct a large-scale evaluation of state-of-the-art LLMs on multiple tool-calling benchmarksBFCL v3, TauBench, Tau2Bench, and AceBenchand introduce MAVEN (Math & Physics Adversarial Verification & Evaluation Network), a new out of distribution (OOD) benchmark designed to stress-test multi-step reasoning through explicit verification and adversarial task composition. Our results show that most current models achieve below 50% accuracy on MAVEN, revealing a significant generalization gap across tool-use settings. To address this, we present the CoreThink Agentic Reasoner, a framework that augments LLMs with a lightweight symbolic reasoning layer for structured decomposition and adaptive tool orchestration. Without additional training, it generalizes across all benchmarks, achieving state-of-the-art performance with 530% improvements over existing baselines at roughly one-tenth the computational cost.