LogoAgentHunter
  • Submit
  • Industries
  • Categories
  • Agency
Logo
LogoAgentHunter

Discover, Compare, and Leverage the Best AI Agents

Featured On

Featured on yo.directory
yo.directory
Featured on yo.directory
Featured on Startup Fame
Startup Fame
Featured on Startup Fame
AIStage
Listed on AIStage
Sprunkid
Featured on Sprunkid
Featured on Twelve Tools
Twelve Tools
Featured on Twelve Tools
Listed on Turbo0
Turbo0
Listed on Turbo0
Featured on Product Hunt
Product Hunt
Featured on Product Hunt
Game Sprunki
Featured on Game Sprunki
AI Toolz Dir
Featured on AI Toolz Dir
Featured on Microlaunch
Microlaunch
Featured on Microlaunch
Featured on Fazier
Fazier
Featured on Fazier
Featured on Techbase Directory
Techbase Directory
Featured on Techbase Directory
backlinkdirs
Featured on Backlink Dirs
Featured on SideProjectors
SideProjectors
Featured on SideProjectors
Submit AI Tools
Featured on Submit AI Tools
AI Hunt
Featured on AI Hunt
Featured on Dang.ai
Dang.ai
Featured on Dang.ai
Featured on AI Finder
AI Finder
Featured on AI Finder
Featured on LaunchIgniter
LaunchIgniter
Featured on LaunchIgniter
Imglab
Featured on Imglab
AI138
Featured on AI138
600.tools
Featured on 600.tools
Featured Tool
Featured on Featured Tool
Dirs.cc
Featured on Dirs.cc
Ant Directory
Featured on Ant Directory
Featured on MagicBox.tools
MagicBox.tools
Featured on MagicBox.tools
Featured on Code.market
Code.market
Featured on Code.market
Featured on LaunchBoard
LaunchBoard
Featured on LaunchBoard
Genify
Featured on Genify
Copyright © 2025 All Rights Reserved.
Product
  • AI Agents Directory
  • AI Agent Glossary
  • Industries
  • Categories
Resources
  • AI Agentic Workflows
  • Blog
  • News
  • Submit
  • Coummunity
  • Ebooks
Company
  • About Us
  • Privacy Policy
  • Terms of Service
  • Sitemap
Friend Links
  • AI Music API
  • ImaginePro AI
  • Dog Names
  • Readdit Analytics
Back to News List

Former DeepSeeker and team unveil RAGEN for training more reliable AI agents

April 23, 2025•Carl Franzen•Original Link•2 minutes
AI
MachineLearning
ReinforcementLearning

RAGEN represents both a technical and conceptual advancement toward autonomous AI agents capable of reasoning and adaptation.

April 23, 2025 — A collaborative team from Northwestern University, Microsoft, Stanford, and the University of Washington, including former DeepSeek researcher Zihan Wang, has developed RAGEN, a new method for training and evaluating AI agents. The system aims to address the brittleness of current AI agents, making them more reliable for real-world applications.

Credit: VentureBeat made with Midjourney

The Challenge: AI Agents Stuck in "Corporate Purgatory"

Despite 2025 being dubbed the "year of AI agents," most implementations remain experimental, according to a VentureBeat poll. RAGEN tackles this by focusing on multi-turn, interactive settings where agents must adapt, remember, and reason under uncertainty.

How RAGEN Works

Built on the StarPO (State-Thinking-Actions-Reward Policy Optimization) framework, RAGEN emphasizes learning through experience rather than memorization. Key features include:

  • Rollout stage: LLMs generate complete interaction sequences guided by reasoning.
  • Update stage: Models are optimized using normalized cumulative rewards.

The team tested RAGEN using Alibaba’s Qwen models (1.5 and 2.5), chosen for their open weights and instruction-following capabilities.

The "Echo Trap" Problem

Wang highlighted a core issue in a widely shared X thread: RL systems often reward shortcuts, leading to repetitive behaviors and degraded performance. This "Echo Trap" manifests as reward variance cliffs and disappearing reasoning traces.

Testing Environments

RAGEN evaluates agents across three symbolic tasks:

  1. Bandit: Single-turn, stochastic risk-reward reasoning.
  2. Sokoban: Multi-turn, deterministic puzzle-solving.
  3. Frozen Lake: Stochastic, multi-turn adaptive planning.

Stabilizing Training with StarPO-S

To combat training collapse, the team introduced StarPO-S, which incorporates:

  1. Uncertainty-based rollout filtering.
  2. KL penalty removal for greater exploration.
  3. Asymmetric PPO clipping to amplify high-reward trajectories.

Key Insights for Effective Agent Training

  • Task diversity improves generalization.
  • Interaction granularity enables meaningful planning.
  • Rollout freshness aligns training data with current policies.

An interactive demo visualizes agent rollouts, including intermediate reasoning steps.

Open-Source Release

RAGEN and its frameworks are now available on GitHub, though licensing details are pending.

Unanswered Questions

  • How transferable is RAGEN beyond symbolic tasks?
  • Can reasoning be sustained over longer horizons?
  • What are the implications for enterprise adoption?

RAGEN represents a significant step toward autonomous, reasoning-capable AI agents, though real-world deployment challenges remain.

Related News

August 18, 2025•Kaydence Shum

Lenovo Wins Frost Sullivan 2025 Asia-Pacific AI Services Leadership Award

Lenovo earns Frost Sullivan's 2025 Asia-Pacific AI Services Customer Value Leadership Recognition for its value-driven innovation and real-world AI impact.

AI
Lenovo
Asia-Pacific
August 18, 2025•Unknown

Baidu Wenku GenFlow 2.0 Revolutionizes AI Agents with Multi-Agent Architecture

Baidu Wenku's GenFlow 2.0 introduces a multi-agent system for parallel task processing, integrating with Cangzhou OS to enhance efficiency and redefine AI workflows.

AI
MultiAgent
Baidu

About the Author

Dr. Lisa Kim

Dr. Lisa Kim

AI Ethics Researcher

Leading expert in AI ethics and responsible AI development with 13 years of research experience. Former member of Microsoft AI Ethics Committee, now provides consulting for multiple international AI governance organizations. Regularly contributes AI ethics articles to top-tier journals like Nature and Science.

Expertise

AI Ethics
Algorithmic Fairness
AI Governance
Responsible AI
Experience
13 years
Publications
95+
Credentials
2
LinkedInResearchGate

Agent Newsletter

Get Agentic Newsletter Today

Subscribe to our newsletter for the latest news and updates