Reinforcement learning (RL) has shown promise for training agentic models that engage in dynamic, multi-turn interactions. However, the value of such agents ultimately depends on their ability to assist users with diverse goals, preferences, and evolving interaction patterns. We introduce **UserRL**, a reproducible and extensible testbed for training and evaluating user-centric agents through standardized gym environments paired with simulated users. UserRL supports flexible reward specification by decoupling turn-level reward shaping from trajectory-level scoring, enabling controlled studies of credit assignment in multi-turn RL. Across Qwen3 models, our experiments show that SFT cold start provides an effective initialization for user-centric RL; trajectory-level reward-to-go scoring improves both effectiveness and interaction efficiency; and stronger simulated users improve training, while open-source simulators remain a cost-effective alternative with positive cross-simulator and preliminary real-user transfer. These results highlight reward design and user simulation as key to user-centric agent training, positioning UserRL as a practical foundation for interactive agent RL.

Large Language Models (LLMs) are typically adapted to downstream tasks through Supervised Fine-Tuning (SFT), where Cross Entropy (CE) loss is used to maximize the likelihood of ground-truth tokens. While effective for aligning with observed data and establishing strong domain grounding, SFT alone often limits the model’s ability to explore alternative plausible outputs, potentially narrowing its decision space. Conversely, Reinforcement Learning (RL) enables reward-driven exploration and refinement but relies heavily on a well-aligned initialization to achieve stable and sample-efficient training. In real-world deployments, practitioners frequently grapple with the question: How much SFT is sufficient, and when should RL take over? This work investigates the complementary trade-off between SFT and RL approaches such as GRPO, focusing on how both stages jointly influence model exploration and exploitation. We provide a quantitative analysis across multiple domains to understand how varying the extent of SFT influences a model’s exploratory behavior and downstream RL effectiveness. Our findings aim to guide in determining optimal SFT checkpoints (CKPTs) that balance domain alignment with exploration capacity, thereby maximizing final domain performance while minimizing overfitting after the joint post-training. Our data and code are open source at https://github.com/Shirley-Kokane/sft2rl.

Autonomous agents powered by large language models (LLMs) have attracted significant research interest. However, the open-source community faces many challenges in developing specialized models for agent tasks, driven by the scarcity of high-quality agent datasets and the absence of standard protocols in this area. We introduce xLAM, a series of large action models designed for AI agent tasks. The xLAM series includes five models with both dense and mixture-of-expert architectures, ranging from 1B to 8x22B parameters, trained using a scalable, flexible pipeline that unifies, augments, and synthesizes diverse datasets to enhance AI agents’ generalizability and performance across varied environments. Our experimental results demonstrate that xLAM consistently delivers exceptional performance across multiple agent ability benchmarks, notably securing the 1st position on the Berkeley Function-Calling Leaderboard, outperforming GPT-4, Claude-3, and many other models in terms of tool use. By releasing the xLAM series, we aim to advance the performance of open-source LLMs for autonomous AI agents, potentially accelerating progress and democratizing access to high-performance models for agent tasks.

Large Action Models (LAMs) for AI Agents offer incredible potential but face challenges due to the need for high-quality training data, especially for multi-steps tasks that involve planning, executing tool calls, and responding to feedback. To address these issues, we present LAM SIMULATOR, a comprehensive framework designed for online exploration of agentic tasks with high-quality feedback. Our framework features a dynamic task query generator, an extensive collection of tools, and an interactive environment where Large Language Model (LLM) Agents can call tools and receive real-time feedback. This setup enables LLM Agents to explore and solve tasks autonomously, facilitating the discovery of multiple approaches to tackle any given task. The resulting action trajectory data are then used to create high-quality training datasets for LAMs. Our experiments on popular agentic benchmarks, ToolBench and CRMArena, highlight the effectiveness of LAM SIMULATOR: models trained with self-generated datasets using our framework achieve significant performance gains, up to a 49.3% improvement over their original baselines. LAM SIMULATOR requires minimal human input during dataset creation, highlighting LAM SIMULATOR’s efficiency and effectiveness in speeding up development of AI agents.

Personalization is essential for AI assistants, especially in private AI settings where models are expected to interpret users’ personal data (e.g., conversations, app usage) to understand their background, preferences, and social context. However, due to privacy concerns, existing academic research lacks direct access to such data, making benchmarking difficult. To fill this gap, we propose a synthetic data pipeline that generates realistic user profiles and private documents, enabling the creation of PersonaBench—a benchmark for evaluating models’ ability to understand personal information. Using this benchmark, we assess Retrieval-Augmented Generation (RAG) pipelines on personalized questions and find that current models struggle to accurately extract and answer questions even when provided with the full set of user documents, highlighting the need for improved personalization methods.

We introduce the Principled Reasoning and Acting (PRAct) framework, a novel method for learning and enforcing action principles from trajectory data. Central to our approach is the use of text gradients from a reflection and optimization engine to derive these action principles. To adapt action principles to specific task requirements, we propose a new optimization framework, Reflective Principle Optimization (RPO). After execution, RPO employs a reflector to critique current action principles and an optimizer to update them accordingly.We investigate the RPO framework under two scenarios: Reward-RPO, which uses environmental rewards for reflection, and Self-RPO, which conducts self-reflection without external rewards. Additionally, we developed two RPO methods, RPO-Traj and RPO-Batch, to adapt to different settings.Experimental results across four environments demonstrate that the PRAct agent, leveraging the RPO framework, can effectively learn and apply action principles to enhance performance.