What is the minimum material needed to win in xiangqi?

These answers restate the rules implemented by our open-source engine, which follows the WXF laws for movement, checkmate and both repetition rulings: a side that forces a repetition by perpetual check or perpetual chase loses rather than drawing. The chase half has not been validated against the official casebook. Where an answer refers to a specific position, that position lives on the linked reference page, where the build verifies it before publishing.

Direct answer

A single chariot forces mate against a bare general — using your own general, which holds the file the enemy would otherwise escape to. So does a lone horse, in every position, and a lone soldier wins unless it has already reached the last rank. A lone cannon cannot even give check. Horse and cannon, two horses, and horse and soldier all win.

The full table

  • Chariot — wins.
  • Horse and cannon — wins.
  • Two horses — wins.
  • Cannon and advisor — wins; the advisor is the screen.
  • Horse and soldier — wins; cannon and soldier — wins 86.8% of positions.
  • Lone horse — wins, in at most 7 moves.
  • Lone soldier — wins, unless it has already reached the last rank.
  • Two soldiers — win 98.4% of positions.
  • Lone cannon — cannot force mate.

Every line is counted position by position on the proven endgame verdicts.

Against a real defence it is harder

These are all against a bare general. Add two advisors and two elephants and the requirements change completely: a lone chariot against a full defence is a standard draw, and horse-and-cannon is often held too. Count the defence separately from the attackers, always.

Related questions

Why can a lone cannon not mate?

It needs a screen to check at all, and an empty palace offers none.

Does the general help?

Yes, essentially. It denies the enemy general a whole file, and several of these mates do not work without it.

Try it on a real board

Reading an answer is one thing; playing it is another. Free, instant, and nothing to install.