Papers
arxiv:2601.03872

Atlas: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning

Published on Jan 7
ยท Submitted by
Jinyang Wu
on Jan 8
#3 Paper of the day
Authors:
,
,
,
,
,
,

Abstract

ATLAS is a dual-path framework that dynamically selects optimal model-tool combinations for cross-domain reasoning through cluster-based routing and reinforcement learning-based multi-step routing, achieving superior performance on complex reasoning tasks.

AI-generated summary

The integration of large language models (LLMs) with external tools has significantly expanded the capabilities of AI agents. However, as the diversity of both LLMs and tools increases, selecting the optimal model-tool combination becomes a high-dimensional optimization challenge. Existing approaches often rely on a single model or fixed tool-calling logic, failing to exploit the performance variations across heterogeneous model-tool pairs. In this paper, we present ATLAS (Adaptive Tool-LLM Alignment and Synergistic Invocation), a dual-path framework for dynamic tool usage in cross-domain complex reasoning. ATLAS operates via a dual-path approach: (1) training-free cluster-based routing that exploits empirical priors for domain-specific alignment, and (2) RL-based multi-step routing that explores autonomous trajectories for out-of-distribution generalization. Extensive experiments across 15 benchmarks demonstrate that our method outperforms closed-source models like GPT-4o, surpassing existing routing methods on both in-distribution (+10.1%) and out-of-distribution (+13.1%) tasks. Furthermore, our framework shows significant gains in visual reasoning by orchestrating specialized multi-modal tools.

Community

Paper author Paper submitter

๐Ÿš€ [New Paper] Atlas: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning

The growing diversity of LLMs and external tools presents a significant challenge: how to select the optimal model-tool combination for complex reasoning tasks. Existing methods often fall short by relying on single models or fixed tool-calling logic. ATLAS (Adaptive Tool-LLM Alignment and Synergistic Invocation) addresses this by introducing a dual-path framework for dynamic model-tool alignment and invocation across multiple domains.

โœจ The Core Intuition:

ATLAS employs a dual-path approach to achieve dynamic model-tool alignment and invocation:

1๏ธโƒฃ Training-Free Cluster-Based Routing: This path leverages empirical priors for domain-specific alignment, efficiently guiding the model-tool selection process.

2๏ธโƒฃ RL-Based Multi-Step Routing: This path explores autonomous trajectories to achieve strong generalization, particularly for out-of-distribution tasks.

๐Ÿ“ˆ Highlights:

Superior Performance: ATLAS significantly outperforms closed-source models like GPT-4o and existing routing methods, achieving +10.1% on in-distribution tasks and +13.1% on out-of-distribution tasks across 15 benchmarks.

Enhanced Visual Reasoning: The framework demonstrates substantial improvements in visual reasoning by effectively orchestrating specialized multi-modal tools.

Adaptive Orchestration: ATLAS learns to assess its internal state and dynamically invoke external resources, internalizing the alignment between domains and tool utilization.

Robust and Generalizable: The design ensures that the routing policy effectively captures expertise distribution, making it robust and generalizable even as tools and models evolve.

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend

Sign up or log in to comment

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2601.03872 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2601.03872 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2601.03872 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.