mlsanityData to 14 Oct 2026

SCAD: Structured Credit Assignment and Distillation for Long-Horizon Agents

arXiv 2610.03372 ↗First posted 2 Oct 2026cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Training long-horizon agents to solve complex tasks requires effective supervision over extended interaction sequences. However, sparse terminal rewards obscure intermediate contributions, while on-policy distillation can lose informative teacher guidance as student-generated histories grow. To address this problem, we introduce SCAD, which organizes interactions into planning and bounded subtask execution, distills execution in local contexts, and refines planning credit through cross-rollout subtask prefix trees, with planning receiving full terminal credit and execution receiving positive terminal credit and teacher guidance. Across all evaluated benchmarks, SCAD improves macro-average accuracy over the strongest training baseline by 4.48 percentage points for text tasks and 4.19 points for multimodal tasks. SCAD effectively combines outcome-based credit assignment with teacher-guided distillation to improve planning and execution in long-horizon agents.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 52 of its 59 references that are in the map. First posted on arXiv 2 Oct 2026.

Builds on

52 of 59 references.

References · 52

  1. 2017Proximal Policy Optimization Algorithmscited by 15,215
  2. 2015Distilling the Knowledge in a Neural Networkcited by 10,324
  3. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  4. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  5. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,754
  6. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,955
  7. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 2,057
  8. 2025Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learningcited by 1,444
  9. 2020Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Stepscited by 1,282
  10. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,261
  11. 2022When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memoriescited by 1,128
  12. 2023GAIA: a benchmark for General AI Assistantscited by 1,128
38 more
  1. 2022Measuring and Narrowing the Compositionality Gap in Language Modelscited by 1,075
  2. 2023On-Policy Distillation of Language Models: Learning from Self-Generated Mistakescited by 939
  3. 2021MuSiQue: Multihop Questions via Single-hop Question Compositioncited by 916
  4. 2025Group-in-Group Policy Optimization for LLM Agent Trainingcited by 497
  5. 2026Reinforcement Learning via Self-Distillationcited by 332
  6. 2026Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipecited by 299
  7. 2023Can Pre-trained Vision and Language Models Answer Visual Information-Seeking Questions?cited by 207
  8. 2025BrowseComp-Plus: A More Fair and Transparent Evaluation Benchmark of Deep-Research Agentcited by 200
  9. 2025WebDancer: Towards Autonomous Information Seeking Agencycited by 188
  10. 2024VinePPO: Refining Credit Assignment in RL Training of LLMscited by 183
  11. 2025WebWalker: Benchmarking LLMs in Web Traversalcited by 175
  12. 2018RUDDER: Return Decomposition for Delayed Rewardscited by 173
  13. 2019Mastering Complex Control in MOBA Games with Deep Reinforcement Learningcited by 161
  14. 2025Scaling Long-Horizon LLM Agent via Context-Foldingcited by 148
  15. 2024MMSearch: Benchmarking the Potential of Large Models as Multi-modal Search Enginescited by 101
  16. 2019Distilling Policy Distillationcited by 84
  17. 2025DistiLLM-2: A Contrastive Approach Boosts the Distillation of LLMscited by 84
  18. 2018Kickstarting Deep Reinforcement Learningcited by 81
  19. 2025TreeRL: LLM Reinforcement Learning with On-Policy Tree Searchcited by 80
  20. 2025Tree Search for LLM Agent Reinforcement Learningcited by 79
  21. 2025In-the-Flow Agentic System Optimization for Effective Planning and Tool Usecited by 72
  22. 2025SimpleVQA: Multimodal Factuality Evaluation for Multimodal Large Language Modelscited by 66
  23. 2026Hierarchy-of-Groups Policy Optimization for Long-Horizon Agentic Taskscited by 49
  24. 2026TCOD: Exploring Temporal Curriculum in On-Policy Distillation for Multi-turn Autonomous Agentscited by 47
  25. 2025MMSearch-Plus: Benchmarking Provenance-Aware Search for Multimodal Browsing Agentscited by 38
  26. 2026Vision-DeepResearch Benchmark: Rethinking Visual and Textual Search for Multimodal Large Language Modelscited by 34
  27. 2025VOLD: Reasoning Transfer from LLMs to Vision-Language Models via On-Policy Distillationcited by 31
  28. 2025Seeking and Updating with Live Visual Knowledgecited by 24
  29. 2025IterResearch: Rethinking Long-Horizon Agents with Interaction Scalingcited by 23
  30. 2026Prefix Teach, Suffix Fade: Local Teachability Collapse in Strong-to-Weak On-Policy Distillationcited by 16
  31. 2026On-policy Distillation with Verifiable Rewardcited by 15
  32. 2026VisBrowse-Bench: Benchmarking Visual-Native Search for Multimodal Browsing Agentscited by 13
  33. 2026RTMC: Step-Level Credit Assignment via Rollout Treescited by 12
  34. 2026ATOD: Annealed Turn-Aware On-Policy Distillation for Multi-Turn Agentic Taskscited by 11
  35. 2026Sequential Beats Joint: On the Interplay between On-Policy Distillation and RLVRcited by 11
  36. 2026HyperEyes: Dual-Grained Efficiency-Aware Reinforcement Learning for Parallel Multimodal Search Agentscited by 10
  37. 2026VSearcher: Long-Horizon Multimodal Search Agent via Reinforcement Learningcited by 9
  38. 2026TRIAGE: Role-Typed Credit Assignment for Agentic Reinforcement Learningcited by 6

50 most cited of 52 All 52 (JSON)

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Keep Policy Gradient in Charge: Sibling-Guided Credit Distillation for Long-Horizon Tool-Use Agents
  2. 2026TRACE: Turn-level Reward Assignment via Credit Estimation for Long-Horizon Agents
  3. 2026OPID: On-Policy Skill Distillation for Agentic Reinforcement Learning
  4. 2026CHIME: Credit-Aware Hierarchical Memory Evolution for Long-Horizon Agentic Planning
  5. 2026Milestone-Guided Policy Learning for Long-Horizon Language Agents
  6. 2026SkillGate: Training In-Policy Skill Selection in Long-Horizon Agents
  7. 2026TrajLong: Co-Designing Agentic and Long-Context Supervision for Mid-Training
  8. 2026VETTA: Coordinating Turn- and Token-Level Credit Assignment for Multi-Turn LLM Agents