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, 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, -bench, -Bench, and ACEBench, demonstrate that our trained model, AgentScaler, significantly enhances the models’ function-calling capability.

Chain-of-Thought (CoT) has proven effective for eliciting reasoning abilities in large language models, but standard autoregressive decoders treat reasoning and answer tokens uniformly as next-token prediction targets, despite their fundamentally different roles. This homogeneous modeling introduces conflicting inductive biases and significant inference overhead. Latent CoT methods attempt to improve efficiency by removing explicit reasoning tokens, yet often degrade performance, indicating that explicit reasoning supervision remains critical for preserving pretrained reasoning capabilities. To address these challenges, we propose Pre² , a phase-aware reasoning framework that decouples thinking and answering while retaining explicit CoT supervision. Our approach introduces a Guided Latent Space and predict independent pre-thinking and pre-answering latent states to guide the model to generate reasoning tokens and answering tokens. We integrates them with explicit token representations through a hybrid latent–token decoding paradigm. This design increases computation along both width and depth dimensions with minimal parameter overhead, enabling stronger reasoning without sacrificing inference efficiency. Training is conducted exclusively on one in-domain dataset, with direct zero-shot evaluation on six out-of-domain benchmarks, and experiments demonstrate consistent and significant improvements over strong baselines and robust generalization.

Large Language Models (LLMs) often struggle with dynamically changing knowledge and handling unknown static information. Retrieval-Augmented Generation (RAG) is employed to tackle these challenges and has a significant impact on improving LLM performance. In fact, we find that not all questions need to trigger RAG. By retrieving parts of knowledge unknown to the LLM and allowing the LLM to answer the rest, we can effectively reduce both time and computational costs. In our work, we propose a Knowledge Boundary Model (KBM) to express the known/unknown of a given question, and to determine whether a RAG needs to be triggered. Experiments conducted on 11 English and Chinese datasets illustrate that the KBM effectively delineates the knowledge boundary, significantly decreasing the proportion of retrievals required for optimal end-to-end performance. Furthermore, we evaluate the effectiveness of KBM in three complex scenarios: dynamic knowledge, long-tail static knowledge, and multi-hop problems, as well as its functionality as an external LLM plug-in.