# Finding the Move Is Not Winning the Game: XiangqiBench for Closed-Loop Evaluation of LLM Agents

arXiv: https://arxiv.org/abs/2610.02425 · HTML: https://mlsanity.com/paper/2610.02425/

## 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 22 of its 32 references that are in the map.
- First posted on arXiv 1 Oct 2026.

## Placed in

- LLM agents and tool use / [Reinforcement learning for LLM agents](https://mlsanity.com/agents/reinforcement-learning-for-llm-agents/)

## Builds on (most cited first, 22 of 22)

- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [WebArena: A Realistic Web Environment for Building Autonomous Agents](https://mlsanity.com/paper/2307.13854/) (2023)
- [τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains](https://mlsanity.com/paper/2406.12045/) (2024)
- [gpt-oss-120b & gpt-oss-20b Model Card](https://mlsanity.com/paper/2508.10925/) (2025)
- [AgentBench: Evaluating LLMs as Agents](https://mlsanity.com/paper/2308.03688/) (2023)
- [OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments](https://mlsanity.com/paper/2404.07972/) (2024)
- [DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence](https://mlsanity.com/paper/2606.19348/) (2026)
- [DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models](https://mlsanity.com/paper/2512.02556/) (2025)
- [Kimi K2.5: Visual Agentic Intelligence](https://mlsanity.com/paper/2602.02276/) (2026)
- [GameBench: Evaluating Strategic Reasoning Abilities of LLM Agents](https://mlsanity.com/paper/2406.06613/) (2024)
- [LMRL Gym: Benchmarks for Multi-Turn Reinforcement Learning with Language Models](https://mlsanity.com/paper/2311.18232/) (2023)
- [ChessGPT: Bridging Policy Learning and Language Modeling](https://mlsanity.com/paper/2306.09200/) (2023)
- [Chess as a Testbed for Language Model State Tracking](https://mlsanity.com/paper/2102.13249/) (2021)
- [The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence](https://mlsanity.com/paper/2605.26494/) (2026)
- [Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity](https://mlsanity.com/paper/2607.00248/) (2026)
- [Amortized Planning with Large-Scale Transformers: A Case Study on Chess](https://mlsanity.com/paper/2402.04494/) (2024)
- [Mastering Board Games by External and Internal Planning with Language Models](https://mlsanity.com/paper/2412.12119/) (2024)
- [TextArena](https://mlsanity.com/paper/2504.11442/) (2025)
- [GAMEBoT: Transparent Assessment of LLM Reasoning in Games](https://mlsanity.com/paper/2412.13602/) (2024)
- [LLM CHESS: Benchmarking Reasoning and Instruction-Following in LLMs through Chess](https://mlsanity.com/paper/2512.01992/) (2025)
- [Xiangqi-R1: Enhancing Spatial Strategic Reasoning in LLMs for Chinese Chess via Reinforcement Learning](https://mlsanity.com/paper/2507.12215/) (2025)

## Abstract (arXiv)

Static evaluations credit a language model for naming the right move, but an agent must carry a plan through to a verified outcome while an opponent responds. We introduce XiangqiBench, an executable benchmark that measures this difference in Chinese chess: starting from 119 tactical endgames with forced mates supported by engine or checks-only search, an LLM agent must deliver checkmate against an engine defender. An interactive REPL interface separates real moves, state queries, and forward simulation, and we record 8,568 multi-turn trajectories from 12 frontier LLMs under two observation protocols. Three signals that look like competence each overstate closed-loop success. (i) The Conversion Gap: models play the stored reference first move in 26.1% of Sighted trials, yet only 13.9% of these trials end in a win. (ii) The Consistency Gap: the leading model reaches 38.7% pass@3 but only 5.9% pass3, winning all three trials on 7 of the 46 positions it ever wins. (iii) The Simulation Gap: 32.3% of accepted simulation calls stop on an illegal move, and in 49.3% of comparable cases the real defender replies differently from the line the agent simulated; self-authored rollouts check legality but cannot anticipate the opponent. Finding the move is not winning the game: agent evaluations should score closed-loop outcomes and report reliability alongside coverage.