AI Trend Notifier
EN
← wiki

$ cat wiki/concepts/agentic-rl.md

Agentic Reinforcement Learning

conceptupdated 2026-08-20created 2026-05-16

Definition

A paradigm in which an LLM agent learns via RL while interacting with an environment. Instead of a single response, it optimizes sequences of multi-step tool use, planning, and reflection against a reward signal.

Why It Matters

  • Squarely at the center of my interests: the intersection of agents + RL + reasoning
  • A limitation of plain RLHF (evaluating a single response) — agents require sequence-level evaluation
  • One of the core research currents of 2026

State of the Art (2026-08-20)

  • Harnessed agentic RL — the deploy-time harness now participates in training (2026-08-18): Agent Lightning v1.0: Towards Harnessed Agentic RL (arXiv:2608.17528) names the regime and ships a ~3,500-line framework for it. Arbitrary agents connect to training through an LLM endpoint proxy, so the harness owns the environment interaction loop and the trainer observes only LLM request/response pairs. The paper states four other frameworks — verl Uni-Agent, AReaL 2.0, slime, Polar — have adopted the architecture. Reported: Qwen3.5-9B 41.8% → 56.4% on SWE-bench Verified from 6K examples. Consequence: a model post-trained this way is fitted to its harness, which is the mechanism behind ClawGym II: Exploring Black-Box RL on Agent Harness (arXiv:2608.16798)'s finding that such a model stops being harness-agnostic. The harness it used for the 56.4% is not named (source)

  • The competing bet: delete the credit-assignment machinery rather than fix it (2026-08-18): Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements (arXiv:2608.17310) argues evolution strategies suit long-horizon agents better than RL — full-parameter optimization at inference-level GPU memory, and trajectory-level parameter attribution with no reward decomposition across the horizon. Reported: +6.69% over a No-Skill baseline on WebArena-Lite for Qwen-3.5-27B, and its matched baseline beaten in 28 of 36 test-time heuristic-design settings. No head-to-head against agentic RL is reported, and no rollout count is published, so the efficiency claim is memory, not compute (source)

  • Anthropic AAR (2026-04-14): 9 instances of Claude Opus 4.6 worked on an alignment research problem (weak-to-strong supervision) for one week in an agentic RL setup → PGR 97% vs. 23% for human researchers. The first empirical case of RL agents replacing humans on a real-world research problem. → Automated Weak-to-Strong Researcher (AAR)

  • Self-Distilled Agentic RL (2026-05-16 HF Daily #3): Self-Distilled Agentic RL

  • Direct On-Policy Distillation / Direct-OPD (2026-07-06, Tsinghua AIR + ByteDance Seed): Run RL on a cheap weak model, then transfer only the RL-induced policy shift (Δ = log π_T − log π_{T,ref}) to a large model using on-policy distillation. Cuts the cost of deploying RLVR gains at frontier scale. A practical cost lever for the Anthropic AAR-style workload where frontier RL is expensive. → Weak-to-Strong Generalization via Direct On-Policy Distillation

  • Related prior work: ReAct, Reflexion, AutoGPT family (page TBD)

Open Problems

Key Papers

Referenced by

Agent Lightning v1.0: Towards Harnessed Agentic RL (arXiv:2608.17528)Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements (arXiv:2608.17310)Agents (LLM Agents)AI AlignmentAndrej KarpathyAREX: Towards a Recursively Self-Improving Agent for Deep ResearchAutomated Weak-to-Strong Researcher (AAR)Beyond Final Scores: A Systematic Evaluation of Agents for Long-Horizon AI Research and Development (arXiv:2608.13417)Claude Managed AgentsClawGym II: Exploring Black-Box RL on Agent Harness (arXiv:2608.16798)DarwinX: Evolving Agent Harnesses Through Natural Selection (arXiv:2608.07545)DataPrep-Bench: Benchmarking LLMs as Training Data PreparatorsEmbodied AgentsENPIRE: Agentic Robot Policy Self-Improvement in the Real WorldHow Can Rhetoric Reward-Hack AI Reviewers? Dissecting Rhetorical Sensitivity in AI-Based Peer Review (arXiv:2608.08975)Intern-S2-Preview: Scientific Agentic Foundation Model (arXiv:2608.13505)Jim FanLaguna S 2.1LFM2.5-2.6BLiquid AILong-Horizon-Terminal-Bench (LHTB)Mistral AIMolt: A Scalable PyTorch-Native Training Framework for Agentic Reinforcement LearningNVIDIAOpenAIPoolsideReasoning ModelsRing-Zero: Scaling Zero RL to a Trillion Parameters for Emergent ReasoningScaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B AgentSEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement LearningSelf-Distilled Agentic Reinforcement LearningSkill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving SkillsSoftware 3.0Solipsistic Superintelligence is Unlikely to be CooperativeThe Mirage of Optimizing Training Policies: Monotonic Inference Policies as the Real Objective for LLM Reinforcement LearningWeak-to-Strong Generalization via Direct On-Policy DistillationWeekly Synthesis — 2026-W21 (2026-05-11 ~ 2026-05-17)Weekly Synthesis — 2026-W29 (2026-07-13 ~ 2026-07-19)

Sources