The Year in AI Papers
The year/Independent research

Paper 2509.13311

Towards General Agentic Intelligence via Environment Scaling

Runnan Fang, Shihao Cai, Baixuan Li, Jialong Wu, Guangyu Li, Wenbiao Yin, Xinyu Wang, Xiaobin Wang, Liangcai Su, Zhen Zhang, Shibin Wu, Zhengwei Tao, et al.

Published
Sep 2025
Research lab
Independent
Citations
48
GitHub
Not linked

01 In brief

Summary

This paper introduces AgentScaler, a family of models trained to advance general agentic intelligence through systematic environment scaling.

The authors propose a two-stage pipeline: first, they automatically construct diverse, fully simulated environments by collecting over 30,000 APIs, organizing them into domains via community detection, and programmatically materializing tools as executable code grounded in database schemas.

Second, they employ a two-phase agent experience learning strategy, where agents first acquire fundamental tool-calling skills across general domains, then specialize in vertical domains.

Trajectories are generated via simulated human-agent interplay and filtered using a three-stage funnel (validity, environment state alignment, and exact function-call matching).

Experiments on τ-bench, τ2-Bench, and ACEBench show that AgentScaler models (4B, 8B, 30B-A3B) achieve state-of-the-art performance among open-source models under 1T parameters, with AgentScaler-30B-A3B matching trillion-parameter models and approaching closed-source systems.

The paper also analyzes generalization to out-of-distribution benchmarks, stability via pass^k metrics, and the challenge of long-horizon tool calling, noting a negative correlation between tool call count and accuracy.

Limitations include lack of reinforcement learning and validation only up to 30B scale.

02 From the paper

Abstract

Advanced agentic intelligence is a prerequisite for deploying Large Language Models in practical, real-world applications. Diverse real-world APIs demand precise, robust function-calling intelligence, which needs agents to develop these capabilities through interaction in varied environments. The breadth of function-calling competence is closely tied to the diversity of environments in which agents are trained. In this work, we scale up environments as a step towards advancing general agentic intelligence. This gives rise to two central challenges: (i) how to scale environments in a principled manner, and (ii) how to effectively train agentic capabilities from experiences derived through interactions with these environments. To address these, we design a scalable framework that automatically constructs heterogeneous environments that are fully simulated, systematically broadening the space of function-calling scenarios. We further adapt a two-phase agent fine-tuning strategy: first endowing agents with fundamental agentic capabilities, then specializing them for domain-specific contexts. Extensive experiments on agentic benchmarks, tau-bench, tau2-Bench, and ACEBench, demonstrate that our trained model, AgentScaler, significantly enhances the function-calling capability of models.