Reasoning & agents

Trace reasoning from the prompt that started it all through self-consistency, search, verification, retrieval, and the agentic benchmarks — the path from "answer once" to a model that reasons, checks itself, and acts.

• Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
• Self-Consistency Improves Chain of Thought Reasoning in Language Models
• Tree of Thoughts: Deliberate Problem Solving with Large Language Models
• Let's Verify Step by Step
• GenPRM: Generative Process Reward Models
• Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
• ReAct: Synergizing Reasoning and Acting in Language Models
• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
• Toolformer: Language Models Can Teach Themselves to Use Tools
• ToolRL: Reward is All Tool Learning Needs
• Group-in-Group Policy Optimization for LLM Agent Training
• τ-bench: Tool-Agent-User Interaction in Real-World Domains
• SWE-bench: Can Language Models Resolve Real-World GitHub Issues?
• Agent Learning via Early Experience
• Why Language Models Hallucinate
• Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty
• Rewarding Doubt: Calibrated Confidence Expression of LLMs
• Proximal Policy Optimization Algorithms
• Spurious Rewards: Rethinking Training Signals in RLVR
• The Entropy Mechanism of RL for Reasoning Language Models
• MiniMax-M1: Scaling Test-Time Compute with Lightning Attention
• ProRL: Prolonged RL Expands Reasoning Boundaries
• From Hard Refusals to Safe-Completions