Skip to content

Caissa v0.1.0

Latest

Choose a tag to compare

@Aneesh495 Aneesh495 released this 06 Nov 22:12

Caissa is an original UCI engine for standard chess and Chess960. These archives include the native engine, an offline analysis board, and an experimental neural evaluator. The engine defaults to classical evaluation.

Try the two-minute live demo to analyze a position from the release matches and export the result.

  • macOS arm64 and Linux x86-64 packages include a quick start. Add build/caissa to a UCI GUI, or run node dist/analysis-service/server.js with Node.js 24 to open the local board.
  • Both packages passed native tests, 20 perft fixtures, and a live packaged-service test. The source build passed real browser tests at desktop and mobile widths. Distribution builds
  • The strength archive contains 192 independently replayed games at 50 ms per move. Caissa scored −286 relative Elo against unrestricted Stockfish 19, with a 95% interval of [−394, −206]. This includes four Stockfish time forfeits. Against Stockfish 19 at Skill Level 0, it scored +597 relative Elo with a 95% lower endpoint of +442. Conditions, sensitivity and replay

The original research acceptance remains incomplete. The model lost its comparison against classical evaluation, the multicore study remains partial, and some historical raw evidence was retired. Research status