KARL: Knowledge Agents via Reinforcement Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chang, Jonathan D., Drozdov, Andrew, Toshniwal, Shubham, Oertell, Owen, Trott, Alexander, Portes, Jacob, Gupta, Abhay, Koppol, Pallavi, Baheti, Ashutosh, Kulinski, Sean, Zhou, Ivan, Dea, Irene, Opsahl-Ong, Krista, Favreau-Lessard, Simon, Owen, Sean, Ortiz, Jose Javier Gonzalez, Singhvi, Arnav, Andrade, Xabi, Wang, Cindy, Sreenivasan, Kartik, Havens, Sam, Liu, Jialu, DeNiro, Peyton, Sun, Wen, Bendersky, Michael, Frankle, Jonathan
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911489080688640
author Chang, Jonathan D.
Drozdov, Andrew
Toshniwal, Shubham
Oertell, Owen
Trott, Alexander
Portes, Jacob
Gupta, Abhay
Koppol, Pallavi
Baheti, Ashutosh
Kulinski, Sean
Zhou, Ivan
Dea, Irene
Opsahl-Ong, Krista
Favreau-Lessard, Simon
Owen, Sean
Ortiz, Jose Javier Gonzalez
Singhvi, Arnav
Andrade, Xabi
Wang, Cindy
Sreenivasan, Kartik
Havens, Sam
Liu, Jialu
DeNiro, Peyton
Sun, Wen
Bendersky, Michael
Frankle, Jonathan
author_facet Chang, Jonathan D.
Drozdov, Andrew
Toshniwal, Shubham
Oertell, Owen
Trott, Alexander
Portes, Jacob
Gupta, Abhay
Koppol, Pallavi
Baheti, Ashutosh
Kulinski, Sean
Zhou, Ivan
Dea, Irene
Opsahl-Ong, Krista
Favreau-Lessard, Simon
Owen, Sean
Ortiz, Jose Javier Gonzalez
Singhvi, Arnav
Andrade, Xabi
Wang, Cindy
Sreenivasan, Kartik
Havens, Sam
Liu, Jialu
DeNiro, Peyton
Sun, Wen
Bendersky, Michael
Frankle, Jonathan
contents We present a system for training enterprise search agents via reinforcement learning that achieves state-of-the-art performance across a diverse suite of hard-to-verify agentic search tasks. Our work makes four core contributions. First, we introduce KARLBench, a multi-capability evaluation suite spanning six distinct search regimes, including constraint-driven entity search, cross-document report synthesis, tabular numerical reasoning, exhaustive entity retrieval, procedural reasoning over technical documentation, and fact aggregation over internal enterprise notes. Second, we show that models trained across heterogeneous search behaviors generalize substantially better than those optimized for any single benchmark. Third, we develop an agentic synthesis pipeline that employs long-horizon reasoning and tool use to generate diverse, grounded, and high-quality training data, with iterative bootstrapping from increasingly capable models. Fourth, we propose a new post-training paradigm based on iterative large-batch off-policy RL that is sample efficient, robust to train-inference engine discrepancies, and naturally extends to multi-task training with out-of-distribution generalization. Compared to Claude 4.6 and GPT 5.2, KARL is Pareto-optimal on KARLBench across cost-quality and latency-quality trade-offs, including tasks that were out-of-distribution during training. With sufficient test-time compute, it surpasses the strongest closed models. These results show that tailored synthetic data in combination with multi-task reinforcement learning enables cost-efficient and high-performing knowledge agents for grounded reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2603_05218
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle KARL: Knowledge Agents via Reinforcement Learning
Chang, Jonathan D.
Drozdov, Andrew
Toshniwal, Shubham
Oertell, Owen
Trott, Alexander
Portes, Jacob
Gupta, Abhay
Koppol, Pallavi
Baheti, Ashutosh
Kulinski, Sean
Zhou, Ivan
Dea, Irene
Opsahl-Ong, Krista
Favreau-Lessard, Simon
Owen, Sean
Ortiz, Jose Javier Gonzalez
Singhvi, Arnav
Andrade, Xabi
Wang, Cindy
Sreenivasan, Kartik
Havens, Sam
Liu, Jialu
DeNiro, Peyton
Sun, Wen
Bendersky, Michael
Frankle, Jonathan
Artificial Intelligence
Machine Learning
We present a system for training enterprise search agents via reinforcement learning that achieves state-of-the-art performance across a diverse suite of hard-to-verify agentic search tasks. Our work makes four core contributions. First, we introduce KARLBench, a multi-capability evaluation suite spanning six distinct search regimes, including constraint-driven entity search, cross-document report synthesis, tabular numerical reasoning, exhaustive entity retrieval, procedural reasoning over technical documentation, and fact aggregation over internal enterprise notes. Second, we show that models trained across heterogeneous search behaviors generalize substantially better than those optimized for any single benchmark. Third, we develop an agentic synthesis pipeline that employs long-horizon reasoning and tool use to generate diverse, grounded, and high-quality training data, with iterative bootstrapping from increasingly capable models. Fourth, we propose a new post-training paradigm based on iterative large-batch off-policy RL that is sample efficient, robust to train-inference engine discrepancies, and naturally extends to multi-task training with out-of-distribution generalization. Compared to Claude 4.6 and GPT 5.2, KARL is Pareto-optimal on KARLBench across cost-quality and latency-quality trade-offs, including tasks that were out-of-distribution during training. With sufficient test-time compute, it surpasses the strongest closed models. These results show that tailored synthetic data in combination with multi-task reinforcement learning enables cost-efficient and high-performing knowledge agents for grounded reasoning.
title KARL: Knowledge Agents via Reinforcement Learning
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2603.05218