Beyond Reward Engineering: A Data Recipe for Long-Context Reinforcement Learning

arXiv:2606.18831 · cs.CL, cs.AI · Submitted 2026-06-17 · Read on arXiv

cs.CL, cs.AI

Submitted: 2026-06-17

Updated: 2026-10-13

Code: https://github.com/gkamradt/LLMTest_NeedleInAHaystack

Terminology

• Reinforcement Learning — 2× in this paper · explained in 4 episodes
• Agentic Task — 1× in this paper · explained in 1 episode
• Agentic Tasks — 1× in this paper · explained in 1 episode
• Benchmark — 1× in this paper · explained in 2 episodes
• GRPO — 1× in this paper · explained in 6 episodes
• Reinforcement Learning (RL) — 1× in this paper · explained in 25 episodes

Sources

• LongRLVR: Long-Context Reinforcement Learning Requires Verifiable Context Rewards
• AgentCPM-Explore: Realizing Long-Horizon Deep Exploration for Edge-Scale Agents
• Evidence-Augmented Policy Optimization with Reward Co-Evolution for Long-Context Reasoning
• ALR$^2$: A Retrieve-then-Reason Framework for Long-context Question Answering
• Large Language Models Can Self-Improve in Long-context Reasoning
• Evaluating Test-Time Scaling of General LLM Agents
• LongReason: A Synthetic Long-Context Reasoning Benchmark via Context Expansion
• WebExplorer: Explore and Evolve for Training Long-Horizon Web Agents
• DeepDive: Advancing Deep Search Agents with Knowledge Graphs and Multi-Turn RL
• OpenAI o1 System Card
• Incentivizing In-depth Reasoning over Long Contexts with Process Advantage Shaping
• LongAct: Harnessing Intrinsic Activation Patterns for Long-Context Reinforcement Learning
• LongR: Unleashing Long-Context Reasoning via Reinforcement Learning with Dense Utility Rewards
• UI-TARS: Pioneering Automated GUI Interaction with Native Agents
• Qwen3 Technical Report
• DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
• QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management
• Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism
• ACC: Compiling Agent Trajectories for Long-Context Training
• Michelangelo: Long Context Evaluations Beyond Haystacks via Latent Structure Queries

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering