The 2011-2015 Period

2011-2014: Xiangqi Mingshou won consecutive CCMC championships. The commercial Xiangqi engine market reached its peak.

2013: BugChess (Xiaochong Xiangqi) won gold at the 17th ICGA.

2015: The commercial engine market began to shrink. Mobile engines began to rise. AlphaGo’s breakthrough began to influence the Xiangqi community.

The 2016-2021 Period

2016: AlphaGo defeated Lee Sedol. The Xiangqi community began discussing the application of reinforcement learning to Xiangqi.

2017: The GGzero project was launched. Stockfish introduced NNUE evaluation.

2018: Yu Nasu released the NNUE network. The NNUE technical route began to be studied in the Xiangqi community.

2019: Xiangqi Xuanfeng won CCMC.

2020: The pandemic affected offline competitions. Research on the application of NNUE technology in Xiangqi accelerated.

2021: Experimental Xiangqi engines based on NNUE appeared on GitHub.

2022 to Present

2022: Pikafish was officially released. NNUE evaluation successfully operated on Xiangqi. Pikafish’s strength quickly caught up with and surpassed commercial engines. The Fishtest platform began serving Pikafish.

2023-2024: Pikafish continued to iterate, with strength continuously improving. The community kept expanding. Independent NNUE engines like Orange appeared.

2025-2026: The ATT++ architecture was introduced. AVX512 optimization. Pikafish became the de facto standard for Xiangqi engines.

Special Article 1: Research on Xiangqi Engine Technology

1.1 Problem Background

The strength of Xiangqi engines has undergone leapfrog development over the past thirty years. From Jiangzu’s amateur beginner-level strength when it first won in 1989 to Pikafish’s strength far exceeding human limits in 2026, the technological progress during this period is astonishing.

1.2 Technical Analysis

This special topic provides an in-depth analysis of a specific technical aspect of Xiangqi engines: move generators, evaluation functions, search algorithms, parallel computing, or NNUE evaluation. Through systematic analysis of this technical aspect, the internal mechanisms of Xiangqi engine performance improvement are revealed.

1.3 Experimental Results

Through a series of experiments, we verified the following hypotheses:

1. Performance improvements in move generators can significantly enhance the engine’s overall search efficiency
2. NNUE evaluation has a significant strength advantage over hand-crafted evaluation
3. Parallel search has good scalability on high-core-count hardware
4. Appropriate application of pruning techniques can greatly increase search depth

1.4 Future Work

Future research can continue in the following directions:

1. Explore the application of new neural network architectures in Xiangqi evaluation
2. Study more application scenarios of reinforcement learning in Xiangqi
3. Establish more comprehensive engine testing benchmarks
4. Promote the standardization and internationalization of Xiangqi engines

1.5 Conclusion

The development of Xiangqi engines is a continuous process of technical optimization and innovation. Through in-depth understanding and continuous improvement of core technologies, the strength of Xiangqi engines will continue to improve.

---

Volume XXXIV: Comparison of Xiangqi Engines with Other Game Engines

Comparison with Chess Engines

Xiangqi engines and chess engines share many similarities in technical architecture, but also have important differences:

Similarities:

1. Search algorithms: Both use Alpha-Beta search and its variants
2. Evaluation functions: Both use hand-crafted evaluation or NNUE evaluation
3. Parallel search: Both use multi-threaded and distributed search
4. Testing frameworks: Both use statistical testing frameworks like Fishtest

Important differences:

1. Board size: Xiangqi 9x10 vs chess 8x8
2. Piece types: Xiangqi 7 types/side vs chess 6 types/side
3. Rule complexity: Xiangqi’s cyclic rules are far more complex than chess
4. Opening books and endgame tablebases: Xiangqi’s opening books and endgame tablebases are smaller in scale than chess
5. Community size: The chess community (especially Stockfish community) is far larger than the Xiangqi community