/

A global community for AI builders, researchers, and founders.

Explore

Updates
Leaderboard
Blog
Workshops

Community

Chapters
Events
Guidelines

Stay Updated

Get updates on events, new chapters, and AI insights.

By subscribing, you agree to our Privacy Policy.

© 2026 AI Socratic. All rights reserved.

Brand Contact Privacy Terms

Search

Search across updates, events, members, and blog posts

Scaling Agentic Search to 20+ Turns via GRPO - Updates | AI Socratic

Back to updates

Research 2025-12-24 @return of the research era ꙮ

Scaling Agentic Search to 20+ Turns via GRPO

Tencent researchers used GRPO to solve "horizon collapse" in 8B models, scaling agentic search to 20+ turns and tripling accuracy by focusing on global rewards rather than token-matching.

#grpo#agentic-search#qwen#reinforcement-learning#optimization

Comments (0)

Sign in to join the discussion.

No comments yet. Be the first to comment!

A global community for AI builders, researchers, and founders.

Explore

Updates
Leaderboard
Blog
Workshops

Community

Chapters
Events
Guidelines

Stay Updated

Get updates on events, new chapters, and AI insights.

By subscribing, you agree to our Privacy Policy.

© 2026 AI Socratic. All rights reserved.

Brand Contact Privacy Terms