# Multi-Turn Reinforcement Learning for Tool-Calling Agents with Iterative Reward Calibration

arXiv: https://arxiv.org/abs/2604.02869 · HTML: https://mlsanity.com/paper/2604.02869/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use.
- Builds on 9 of its 15 references that are in the map.
- First posted on arXiv 3 Apr 2026.

## Placed in

- LLM agents and tool use / [Reinforcement learning for LLM agents](https://mlsanity.com/agents/reinforcement-learning-for-llm-agents/)
- Deep reinforcement learning methods / [RL training of language and agent models](https://mlsanity.com/deeprl/rl-training-of-language-and-agent-models/)

## Builds on (most cited first, 9 of 9)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Toolformer: Language Models Can Teach Themselves to Use Tools](https://mlsanity.com/paper/2302.04761/) (2023)
- [τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains](https://mlsanity.com/paper/2406.12045/) (2024)
- [GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization](https://mlsanity.com/paper/2601.05242/) (2026)
- [WebAgent-R1: Training Web Agents via End-to-End Multi-Turn Reinforcement Learning](https://mlsanity.com/paper/2505.16421/) (2025)
- [Turn-PPO: Turn-Level Advantage Estimation with PPO for Improved Multi-Turn RL in Agentic LLMs](https://mlsanity.com/paper/2512.17008/) (2025)
- [Proximity-Based Multi-Turn Optimization: Practical Credit Assignment for LLM Agent Training](https://mlsanity.com/paper/2602.19225/) (2026)
- [Empowering Multi-Turn Tool-Integrated Agentic Reasoning with Group Turn Policy Optimization](https://mlsanity.com/paper/2511.14846/) (2025)

## Built on by (most cited first, 7 of 7)

- [Exploring Agentic Tool-Calling Decisions via Uncertainty-Aligned Reinforcement Learning](https://mlsanity.com/paper/2606.06976/) (2026)
- [Mach-Mind-4-Flash Technical Report](https://mlsanity.com/paper/2607.09375/) (2026)
- [RLVP: Penalize the Path, Reward the Outcome](https://mlsanity.com/paper/2607.07435/) (2026)
- [SkillGate: Training In-Policy Skill Selection in Long-Horizon Agents](https://mlsanity.com/paper/2608.18852/) (2026)
- [SpeechGym: An Audio-Native Gym for Training Voice Agents via Reinforcement Learning](https://mlsanity.com/paper/2608.26432/) (2026)
- [Coverage, Not Targeting: A Structural Regime in Multi-Turn Agent Credit Assignment](https://mlsanity.com/paper/2609.02417/) (2026)
- [Entropy Pacing Policy Optimization for Multi-Task Agentic Reinforcement Learning](https://mlsanity.com/paper/2607.07178/) (2026)

## Abstract (arXiv)

Training tool-calling agents with reinforcement learning on multi-turn tasks remains challenging due to sparse outcome rewards and difficult credit assignment across conversation turns. We present the first application of MT-GRPO (Multi-Turn Group Relative Policy Optimization) combined with GTPO (Generalized Token-level Policy Optimization) for training a tool-calling agent on realistic customer service tasks with an LLM-based user simulator. Through systematic analysis of training rollouts, we discover that naively designed dense per-turn rewards degrade performance by up to 14 percentage points due to misalignment between reward discriminativeness and advantage direction. We introduce Iterative Reward Calibration, a methodology for designing per-turn rewards using empirical discriminative analysis of rollout data, and show that our GTPO hybrid advantage formulation eliminates the advantage misalignment problem. Applied to the Tau-Bench airline benchmark, our approach improves Qwen3.5-4B from 63.8 percent to 66.7 percent (+2.9pp) and Qwen3-30B-A3B from 58.0 percent to 69.5 percent (+11.5pp) -- with the trained 4B model exceeding GPT-4.1 (49.4 percent) and GPT-4o (42.8 percent) despite being 50 times smaller, and the 30.5B MoE model approaching Claude Sonnet 4.5 (70.0 percent). To our knowledge, these are the first published RL training results on Tau-Bench. We release our code, reward calibration analysis, and training recipes.