mlsanityData to 14 Oct 2026

Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards

arXiv 2610.18868 ↗First posted 13 Oct 2026cs.CL cs.AI cs.CVcited by 0

On the map: LLM agents and tool use → Reinforcement learning for LLM agents also fits: Agent skill learning and evolution

Abstract (arXiv)

Long-horizon agents have drawn growing attention for their ability to work autonomously -- from hours to days -- on economically valuable tasks, a capability essential to automated discovery and recursive self-improvement (RSI). Yet open-source models still struggle on long-horizon tasks, largely for lack of both training data and a working recipe, and scaling both is hard: long, complex tasks are difficult to curate and verify, and their rewards are sparse. We address both. First, we introduce Tev-Gym, the first long-horizon agent training dataset with verifiable, dense, progressive rewards -- 1,170 verified task environments across 7 domains. It is built on a latent-world-first approach that first constructs a latent world: a complete and correct hidden specification of the task domain, workflow, and expected state after each step. Each task is then progressively rendered from this specification into an agent-facing environment, enabling controllable complexity and a natural progress metric. Second, we turn this metric into a training recipe: via progress-potential reward shaping (PPRS), we convert progress into a dense per-step reward that improves training efficiency and makes otherwise-intractable tasks trainable. Training with this recipe yields Tev-27B, which achieves substantial gains on the Agents' Last Exam computing and math sciences domains (ALE-CM), matching GPT-6 Astra and Claude Opus 5 at Pass@5. Despite training only on ALE-related data, it generalizes to the Terminal-Bench (TB) 3.0 and 4.0 subsets, surpassing GLM 5.2 and Gemini 3.7 Flash, and reaches Claude Opus 4.8-level performance on middle-horizon tasks such as TB 2.1.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use. Builds on 77 of its 84 references that are in the map. First posted on arXiv 13 Oct 2026.

Builds on

77 of 84 references.

Closest ancestors · 12

Same direction, newest first

  1. 2026T1: Terminal Agent Reinforcement Learning for Long-Horizon Tasks10 Sep
  2. 2026The Dark Room in the Reward Channel: Dense Prediction Rewards Collapse GRPO-Trained LLM Agents -- and The Channel, Not the Content, Decides What Works23 Jul
  3. 2026SETA: Scaling Environments for Terminal Agents12 Jul
  4. 2026Verifiable Process Rewards for Agentic Reasoning11 May
  5. 2025AutoForge: Automated Environment Synthesis for Agentic Reinforcement Learning28 Dec
  6. 2025AutoEnv: Automated Environments for Measuring Cross-Environment Agent Learning24 Nov
  7. 2025AgentGym-RL: Training LLM Agents for Long-Horizon Decision Making through Multi-Turn Reinforcement Learning10 Sep
  8. 2025SPA-RL: Reinforcing LLM Agents via Stepwise Progress Attribution27 May
  9. 2025Group-in-Group Policy Optimization for LLM Agent Training16 May
  10. 2025SWEET-RL: Training Multi-Turn LLM Agents on Collaborative Reasoning Tasks19 Mar
  11. 2025Process Reward Models for LLM Agents: Practical Framework and Directions14 Feb
  12. 2024ArCHer: Training Language Model Agents via Hierarchical Multi-Turn RL29 Feb

Other references · 65

  1. 2017Proximal Policy Optimization Algorithmscited by 15,215
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  3. 2019BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehensioncited by 5,893
  4. 2023Let's Verify Step by Stepcited by 3,789
  5. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,373
  6. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,810
  7. 2023WebArena: A Realistic Web Environment for Building Autonomous Agentscited by 2,006
  8. 2024τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domainscited by 1,402
  9. 2024OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environmentscited by 1,195
  10. 2026Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfacescited by 585
  11. 2025Process Reinforcement through Implicit Rewardscited by 504
  12. 2024MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineeringcited by 469
38 more
  1. 2024AppWorld: A Controllable World of Apps and People for Benchmarking Interactive Coding Agentscited by 401
  2. 2025Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domainscited by 366
  3. 2024Training Software Engineering Agents and Verifiers with SWE-Gymcited by 336
  4. 2024TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Taskscited by 326
  5. 2025PaperBench: Evaluating AI's Ability to Replicate AI Researchcited by 321
  6. 2025SWE-smith: Scaling Data for Software Engineering Agentscited by 285
  7. 2025SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolutioncited by 263
  8. 2017Automatic Goal Generation for Reinforcement Learning Agentscited by 259
  9. 2025Skywork Open Reasoner 1 Technical Reportcited by 165
  10. 2025R2E-Gym: Procedural Environments and Hybrid Verifiers for Scaling Open-Weights SWE Agentscited by 158
  11. 2023Self-Alignment with Instruction Backtranslationcited by 151
  12. 2025SWE-rebench: An Automated Pipeline for Task Collection and Decontaminated Evaluation of Software Engineering Agentscited by 142
  13. 2025REASONING GYM: Reasoning Environments for Reinforcement Learning with Verifiable Rewardscited by 130
  14. 2025BREAD: Branched Rollouts from Expert Anchors Bridge SFT & RL for Reasoningcited by 63
  15. 2026EnvScaler: Scaling Tool-Interactive Environments for LLM Agent via Programmatic Synthesiscited by 63
  16. 2025Enigmata: Scaling Logical Reasoning in Large Language Models with Synthetic Verifiable Puzzlescited by 56
  17. 2025Towards General Agentic Intelligence via Environment Scalingcited by 54
  18. 2026Endless Terminals: Scaling RL Environments for Terminal Agentscited by 54
  19. 2026Agent-World: Scaling Real-World Environment Synthesis for Evolving General Agent Intelligencecited by 54
  20. 2024WorkBench: a Benchmark Dataset for Agents in a Realistic Workplace Settingcited by 53
  21. 2025SynLogic: Synthesizing Verifiable Reasoning Data at Scale for Learning Logical Reasoning and Beyondcited by 50
  22. 2025GenEnv: Difficulty-Aligned Co-Evolution Between LLM Agents and Environment Simulatorscited by 42
  23. 2025Kimi-Dev: Agentless Training as Skill Prior for SWE-Agentscited by 41
  24. 2025ALE-Bench: A Benchmark for Long-Horizon Objective-Driven Algorithm Engineeringcited by 39
  25. 2026Rethinking the Trust Region in LLM Reinforcement Learningcited by 37
  26. 2026Tmax: A simple recipe for terminal agentscited by 33
  27. 2026ScaleEnv: Scaling Environment Synthesis from Scratch for Generalist Interactive Tool-Use Agent Trainingcited by 32
  28. 2026Agents' Last Examcited by 31
  29. 2026Long-Horizon-Terminal-Bench: Testing the Limits of Agents on Long-Horizon Terminal Tasks with Dense Reward-Based Gradingcited by 31
  30. 2025Skywork-SWE: Unveiling Data Scaling Laws for Software Engineering in LLMscited by 26
  31. 2026CLI-Gym: Scalable CLI Task Generation via Agentic Environment Inversioncited by 23
  32. 2025SWE-Flow: Synthesizing Software Engineering Data in a Test-Driven Mannercited by 20
  33. 2025InternBootcamp: Boosting LLM Reasoning with Verifiable Task Scalingcited by 20
  34. 2026EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environmentscited by 20
  35. 2026Recursive Synthesis for Long-Horizon Terminal Taskscited by 19
  36. 2026Search-P1: Path-Centric Reward Shaping for Stable and Efficient Agentic RAG Trainingcited by 18
  37. 2025SWE-Dev: Evaluating and Training Autonomous Feature-Driven Software Developmentcited by 16
  38. 2025AgentFrontier: Expanding the Capability Frontier of LLM Agents with ZPD-Guided Data Synthesiscited by 14

50 most cited of 65 All 65 (JSON)

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Long-Horizon-Terminal-Bench 2.0: Automating Diverse Task Construction and Measuring Sustained Agent Progress
  2. 2026SWE-Marathon: Can Agents Autonomously Complete Ultra-Long-Horizon Software Work?
  3. 2026Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent
  4. 2026Beyond Reward Engineering: A Data Recipe for Long-Context Reinforcement Learning
  5. 2026ASCENT: Online Test-Time Training of Long-Horizon Agents via Self-Distillation of Verified Experience
  6. 2026AgentV-RL: Scaling Reward Modeling with Agentic Verifier
  7. 2026Scaling Agentic Data for Long-Horizon Terminal Intelligence
  8. 2026Scalable Environments Drive Generalizable Agents