mlsanityData to 7 Oct 2026

Speculate While You Reason: Teaching Agents to Predict Their Next Tool Call via Joint Agent-Speculator RL

arXiv 2607.25816 ↗First posted 28 Jul 2026cs.AIcited by 1

On the map: LLM reasoning → Reasoning for agents, planning and code

also in 1 more
Abstract (arXiv)

Large language model agents often spend substantial wall-clock time waiting for tool call results. Tool-call speculation can hide this latency by predicting and pre-executing an agent's next tool call if the prediction matches the agent's eventual tool call, but existing speculators are typically separate draft models or cached traces that are poorly aligned with the deployed agent's own behavior. We identify this speculator-agent gap and show that the target agent itself is a strong next-call speculator. This points to a simpler design: unifying the agent and speculator within the same model. In this paper, we introduce the self-speculating agent, a single model that both solves tasks in agent mode and predicts its next tool call from partial trajectories in speculator mode, fully reusing prefix KV cache. To enable this dual-mode agent without degrading performance, we propose a joint agent-speculator reinforcement learning method, which derives speculation targets from the agent's own rollouts and alternates agent and speculator updates. Across agentic search QA and conversational tool-use agentic tasks, our method improves average next tool-call Hit@1 from 44.1 to 61.2 for Qwen3-4B and from 48.9 to 66.3 for Qwen3.5-4B, while preserving agent task success.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 38 of its 44 references that are in the map. First posted on arXiv 28 Jul 2026.

Builds on

38 of 44 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025Reducing Latency of LLM Search Agent via Speculation-based Algorithm-System Co-Design25 Nov
  2. 2025Agentic Reasoning and Tool Integration for LLMs via Reinforcement Learning28 Apr
  3. 2025ReTool: Reinforcement Learning for Strategic Tool Use in LLMs15 Apr

Other references · 35

  1. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  3. 2025Qwen3 Technical Reportcited by 8,195
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  5. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,072
  6. 2022Constitutional AI: Harmlessness from AI Feedbackcited by 2,950
  7. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,678
  8. 2025Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learningcited by 1,371
  9. 2023Gorilla: Large Language Model Connected with Massive APIscited by 1,302
  10. 2024τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domainscited by 1,280
  11. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,193
  12. 2025τ2-Bench: Evaluating Conversational Agents in a Dual-Control Environmentcited by 558
23 more
  1. 2025ToolRL: Reward is All Tool Learning Needscited by 342
  2. 2025R1-Searcher: Incentivizing the Search Capability in LLMs via Reinforcement Learningcited by 322
  3. 2025DeepResearcher: Scaling Deep Research via Reinforcement Learning in Real-world Environmentscited by 268
  4. 2023FireAct: Toward Language Agent Fine-tuningcited by 247
  5. 2024WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learningcited by 243
  6. 2024APIGen: Automated Pipeline for Generating Verifiable and Diverse Function-Calling Datasetscited by 188
  7. 2025ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learningcited by 187
  8. 2025BrowseComp-Plus: A More Fair and Transparent Evaluation Benchmark of Deep-Research Agentcited by 183
  9. 2025ToRL: Scaling Tool-Integrated RLcited by 166
  10. 2025Scaling Long-Horizon LLM Agent via Context-Foldingcited by 136
  11. 2025Reinforcement Learning for Long-Horizon Interactive LLM Agentscited by 103
  12. 2025Agent Lightning: Train ANY AI Agents with Reinforcement Learningcited by 94
  13. 2025Reinforcing Multi-Turn Reasoning in LLM Agents via Fine-Grained Reward Structure and Credit Assignmentcited by 77
  14. 2026How Well Do Agentic Skills Work in the Wild: Benchmarking LLM Skill Usage in Realistic Settingscited by 62
  15. 2025ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestrationcited by 44
  16. 2025Tool-R1: Sample-Efficient Reinforcement Learning for Agentic Tool Usecited by 16
  17. 2026Parallelizing Tool Execution and LLM Generation for Low-Latency Agent Servingcited by 16
  18. 2025RAG-R1: Incentivizing the Search and Reasoning Capabilities of LLMs through Multi-query Parallelismcited by 15
  19. 2025Optimizing Agentic Language Model Inference via Speculative Tool Callscited by 11
  20. 2026DualSpec: Accelerating Deep Research Agents via Dual-Process Action Speculationcited by 6
  21. 2025Collision- and Reachability-Aware Multi-Robot Control with Grounded LLM Plannerscited by 4
  22. 2025HarnessLLM: Automatic Testing Harness Generation via Reinforcement Learningcited by 2
  23. 2026Speculative Interaction Agents: Building Real-Time Agents with Asynchronous I/O and Speculative Tool Callingcited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026TomasuLLM: Out-of-Order Speculative Execution for LLM Agentscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026On Effectiveness and Efficiency of Agentic Tool-calling and RL Training
  2. 2026Self-Evolution for Multi-Turn Tool-Calling Agents via Divergence-Point Preference Learning
  3. 2026Speculate with Memory: Lossless Acceleration for LLM Agents
  4. 2026ForkPilot: Self-Evolving Policy for Retrospective Search in Long-Horizon Agents
  5. 2026CacheRL:Multi-Turn Tool-Calling Agents via Cached Rollouts and Hybrid Reward
  6. 2026MetaForge: A Self-Evolving Multimodal Agent that Retrieves, Adapts, and Forges Tools On Demand ↗
  7. 2026SPORK: Self-Speculative Forking to Accelerate Agentic LLM Inference
  8. 2025Evolving in Tasks: Empowering the Multi-modality Large Language Model as the Computer Use Agent