Chinese ChessAI Capability Benchmark

A standardized benchmark evaluating frontier foundation models and specialized neural engines on discrete spatial reasoning and rules adherence. Featuring edge-native 10M Transformer.

01

LLM Xiangqi Benchmark

Rigorous sandbox and swap-opening pair evaluation assessing frontier models on spatial coherence, legal generation, and tactical solving.

Leaderboard Overview
02

Edge-Native Neural Model

Custom 9.74M Transformer with 32-ply action history attention. Verified past Pikafish UCI_Elo 2000 baseline with zero backend compute cost.

Explore Model Architecture
03

Play on Any Platform

Pure client-side WebAssembly execution with zero server inference cost. Offline native apps for macOS and Android.

Play in Browser Now

Play Online & Client Support

Experience discrete neural chess gameplay across multiple platforms

Web Edition

Zero installation required. Instant client-side inference powered by multithreaded ONNX Runtime Web.

Android Mobile

Ergonomic mobile UI optimized for one-handed play with offline 10M neural model. Full standalone gameplay.

macOS Desktop

Native Compose Desktop application with hardware acceleration and configurable offline search depth.