What Is the Best AI Coach for Learning Chinese Chess?

2026-09-05

AI Chinese chess coach reviewing a Xiangqi position with traditional red and black pieces

How Do AI Xiangqi Coaches Compare on Principle-Based Teaching Versus Engine Move Suggestions?

For players who need to understand why a move works — not only which move an engine prefers — Jenova's Chinese Chess Coach is the strongest conversational option in 2026, while Xiangqi.com, Xiangqi One, and Tiantian Xiangqi (天天象棋) remain stronger for live rated play, puzzle volume, and built-in engine replay. The useful split is teaching versus practice infrastructure.

Key factors that separate effective AI Xiangqi coaching from generic move engines:

✅ Principle-first explanation of screens, tempo, and palace integrity — not just a principal variation ✅ Adaptive calibration from complete beginner through tournament terminology, including WXF notation and endgame result classes ✅ Honest handling of rules that punish chess habits: stalemate is a loss, and perpetual check is a loss ✅ Endgame classification before calculation (full guard versus partial guard) ✅ A separate place to play real games and verify ideas with an engine, which a chat coach does not replace

To compare these tools fairly, it helps to separate Xiangqi study into explanation, practice, and verification — then match each layer to the product that actually owns it.

Why Has AI-Assisted Xiangqi Study Grown in 2026?

AI-assisted Xiangqi study has grown because English-language play sites, open-source engines, and international rules documentation finally sit in one workflow, while high-quality explanation still lags behind move-checking. Xiangqi remains the most popular board game in China, and English-facing platforms now make it realistic to learn without living inside a Chinese-only app.

Scale is no longer the bottleneck. Xiangqi.com reports more than 2 million players from over 100 countries each year, about 36 million games, 9 million puzzle attempts, and 70-plus English-language lessons. Xiangqi One describes the game as having over 500 million players worldwide and ships browser analysis through Pikafish, the leading open-source Xiangqi engine.

The coaching gap is still obvious in community discussions. Players on

Reddit Post

— a sign that play platforms improved faster than teaching platforms. Engines answer “what,” not “when does a cannon lose value” or “which endings are theoretical draws against full guard.”

International rules also matter more than they did a decade ago. The World Xiangqi Federation (WXF), founded in Beijing on 6 April 1993, published unified World Xiangqi Rules in Chinese and English, based on the Asian Xiangqi Federation’s fourth revision. A 2024 technical paper notes that tournament repetition law still has two main official versions, including WXF. Players who only memorize “don’t repeat” get surprised over the board.

What Should You Look for in an AI Xiangqi Coach?

You should look for a coach that teaches Xiangqi as a coordination game — screens, tempo, and palace structure — then sends you somewhere else to play volume. Move quality without a teaching model produces players who copy first-choice engine lines and collapse in original positions.

This article uses an Explanation–Practice–Verification Stack. Weight the layers according to your bottleneck, not according to which product has the flashiest board.

1. False-analogy teaching (explanation). International chess transfers poorly. Horses block at the knee, elephants cannot cross the river, the general stays in the palace, and perpetual check loses. A coach that names these traps early prevents months of unlearning.

2. Named patterns, not anonymous tactics (explanation). 马后炮 (cannon behind horse), 铁门栓 (iron bolt), 卧槽马 (crouching horse), and 白脸将 (throat cut) are study units. If a tool only says “checkmate in two,” it is a puzzle generator, not a coach.

3. Opening principles before trees (explanation). Central Cannon versus Screen Horse (中炮对屏风马) is the deepest system, but beginners need development rules: chariots first, central-file control, cannon-horse coordination.

4. Screen management across the whole game (explanation). Cannons need screens to capture and lose relative value as pieces trade. That curve is unique to Xiangqi and rarely taught by engines.

5. Endgame result classes (explanation). Theoretical win (必胜), theoretical draw (必和), positional win (巧胜), and positional draw (巧和) tell you which endings to aim for. Calculation without classification wastes time.

6. Human games plus engine check (practice and verification). Xiangqi One’s 2026 platform guide treats web and mobile play as complementary. A coach that cannot host rated games is incomplete as a full training environment — and a play site that cannot explain a principle is incomplete as a coach.

Language access is a hidden seventh criterion. Tiantian Xiangqi is built for Chinese-speaking volume. Xiangqi.com supports Simplified Chinese, Traditional Chinese, English, and Vietnamese, plus graphical pieces. A conversational coach should accept WXF, Chinese notation, FEN, and plain language without forcing one format.

How Do Jenova, Xiangqi.com, Xiangqi One, and Tiantian Xiangqi Compare?

They occupy different layers of the stack: Jenova’s Chinese Chess Coach is built for adaptive explanation, Xiangqi.com and Xiangqi One for international play and puzzles, and Tiantian Xiangqi for Chinese-language volume and master video. None of them is a complete substitute for the others.

Jenova Chinese Chess Coach

The coach works in conversation. It diagnoses level from how you talk — “how does the horse move?” versus ECCO codes and engine evals — rather than forcing a placement quiz. It can shift from Socratic prompts when you are learning to direct variation dumps when you want a post-mortem.

Coverage runs from piece movement through opening families (Central Cannon, Same Direction Cannons, Angel Pawn, Elephant Opening), named mates, screen strategy, and mapped endgame results against full guard (士象全). It also translates chess habits into Xiangqi features instead of treating them as defects.

Limitations are structural. There is no clickable board, no ELO pool, and no 60,000-position drill bank. Positions appear as annotated text diagrams. It cannot adjudicate a live WXF repetition dispute the way a tournament platform can. Players still need a play site and, for verification, an engine such as Pikafish.

On Jenova, the free tier includes core access with limited monthly usage. Paid plans start at $20 per month (Plus, 30× free usage), with higher tiers at $50, $100, $200, $500, and $1,000 per month. Cross-session memory is the practical advantage: the coach can remember that you keep blocking your own horse on file 3.

Readers who also study other Chinese table games often keep a Mahjong Tutor in the same rotation. Players who want to read 中炮对屏风马 comments in the original language often pair coaching with Learn Chinese Through Roleplay.

Xiangqi.com

Xiangqi.com is an international play-and-learn site rather than a personal coach. It offers two-player games with timers, computer opponents, puzzles you can also create and share, post-game engine analysis, ELO ratings, and weekly free Swiss tournaments. A comparison page against PlayOK highlights rated and casual play plus multiple computer levels.

Strengths are accessibility and volume. Graphical pieces help non-readers of Chinese. The mobile app adds online multiplayer, computer opponents, thousands of puzzles, and lessons. For an English-speaking beginner who needs games tonight, this is often the first stop.

Limitations: lessons and engine notes are not the same as a coach that tracks your recurring river-horse mistakes across weeks. Computer levels and puzzle quality help tactics more than long-term positional judgment. As of 2026 the core play experience is free, which is a genuine advantage if your budget is zero and your need is opponents.

Xiangqi One

Xiangqi One is a free, ad-free web and mobile platform with 22 AI bots, 1,000-plus puzzles, interactive tutorials, online ELO games in bullet, blitz, and rapid, and built-in Pikafish analysis. Analysis includes move-by-move evaluation, blunder tagging, and opening names.

Its strength is verification. Pikafish in the browser removes the old barrier of installing a UCI engine and a separate graphical interface. Bots give a difficulty ladder without waiting for a human match.

Its limitation is pedagogy. A bot that plays calmly or aggressively still does not ask what your instinct was, or whether a chariot-cannon battery was the right plan against an intact palace. Treat Xiangqi One as a practice-and-check layer, not as a substitute for principle instruction.

Tiantian Xiangqi (天天象棋)

Tiantian Xiangqi is Tencent’s mass-market Chinese app: skill testing, smart game review (智能复盘), a problem set drawn from classic endgames and real games numbering more than 60,000, structured courses, and 30-plus signed-master livestreams and videos. It also includes club rooms, game-record collections, endgame campaign modes, and variants such as 揭棋.

For Chinese-fluent players who want density — puzzles, masters, and a huge opponent pool — it is difficult to match. A 2023 App Store listing put membership at 30 yuan per month, or 24 yuan with auto-renewal; confirm current pricing in the app, because those figures can change.

Limitations for international learners are language and coaching style. Video courses and auto-review still optimize for “what was the better move” more than for a Socratic dialogue about screen risk. If you cannot read the interface, the volume advantage disappears.

Feature / dimension	Xiangqi.com	Jenova Chinese Chess Coach	Xiangqi One	Tiantian Xiangqi
Teaching model	Static lessons, puzzles, post-game engine notes	Adaptive conversation; Socratic or direct; remembers your leaks	Tutorials, bots, daily puzzles	Video courses, 60,000+ drills, master streams
Live human play	Yes — ELO, weekly Swiss	No	Yes — bullet, blitz, rapid ELO	Yes — large Chinese-language pool
Engine / review	Post-game AI analysis	Conceptual analysis; not a click-to-move GUI	Built-in Pikafish	Smart replay (智能复盘)
Endgame framing	Puzzles and articles	必胜 / 必和 / 巧胜 / 巧和 against guard structures	Puzzle-centric	Classic manuals plus 残局闯关
Languages	English, Chinese, Vietnamese; graphical pieces	Conversational; WXF, Chinese notation, FEN	English-forward	Primarily Chinese
Pricing (as of 2026 unless noted)	Free play and lessons	Free limited usage; Plus $20/mo	Free, no ads	About ¥24–30/mo in a 2023 listing
Best for	Rated international games and English onboarding	Principle-based coaching from beginner through tournament study	Free engine-backed practice	High-volume Chinese training

Pikafish sits beside this table rather than inside it. It is a UCI Xiangqi engine with NNUE evaluation, not a teacher. Running it through a GUI is the right verification step after a coaching session — and the wrong first teacher for a newcomer.

Casual link-play against a friend, without ratings or lessons, also exists on sites such as Elephant Chess. That fills a social niche, not a coaching niche.

How Does Screen Management Change What an AI Coach Needs to Teach?

Screen management is Xiangqi’s distinctive strategic skill, so a coach that never talks about screens is teaching a different game than the one you will play. Cannons capture only by jumping exactly one piece — the screen, or 架 — which means every trade, every block, and even your own general can create or destroy attacking lines.

Four coaching points separate useful screen talk from slogan-level advice:

Creating screens. You place a piece so a cannon can capture through it. Beginners hunt captures; stronger players manufacture the 架 two moves earlier.
Destroying screens. Removing the opponent’s hopper often “turns off” a cannon without attacking the cannon itself.
Self-screening. Your own chariot or horse can become the screen that lets the opponent’s cannon hit you. Engines punish this instantly; humans need it named.
The general as a screen and as an attacker. Flying general (对面将) pins whatever sits between the two generals. In endgames the general is an attacking unit, which is why 白脸将 (throat cut) exists.

This is also why cannon value falls and horse value rises as the board empties. Fewer pieces mean fewer screens, so a coach that always recites “cannon equals 4.5, horse equals 4” without a phase trend is giving opening arithmetic for an endgame decision.

A practical drill with Jenova’s Chinese Chess Coach looks like this:

"Set up a midgame where my cannon on file 5 looks active but every screen is my own chariot. Ask me to find the self-screen before you show the engine-style fix."

On Xiangqi One or Xiangqi.com, play the same structure against a bot, then open the engine line and check whether you removed the screen or just moved the cannon. The coach supplies the concept; the play site supplies the reps.

Which Tools Explain Xiangqi Endgame Theory Most Clearly?

Jenova’s Chinese Chess Coach explains endgames as a classification problem first and a calculation problem second, while puzzle platforms explain them as mate races. If you cannot say whether a position is 必胜 or 必和 against the defender’s remaining advisors and elephants, extra tactics problems will not save you.

Xiangqi endgame literature is unusually mapped. The first question is guard structure. Two advisors plus two elephants is full guard (士象全). Typical attacking-force results against that wall, taught as working knowledge rather than trivia:

Chariot alone — generally a draw, and the drawing pattern every student should know
Chariot plus horse — win; the horse restricts the palace, the chariot delivers
Chariot plus cannon — win in most settings
Chariot plus a high (post-river) soldier — win
Horse plus cannon, or horse plus soldier — usually draws
Two horses — positional; some formations win

Against only two advisors or only two elephants, a lone chariot often wins. Losing one guard piece changes the theoretical landscape more than winning a soldier does. Minimum mating knowledge also differs from chess: a chariot can force mate; a single horse or cannon cannot.

Puzzle banks on Xiangqi.com, Xiangqi One, and Tiantian Xiangqi are excellent for 马后炮 and 双鬼拍门 pattern recognition. They are weaker at teaching you which ending to steer toward when you still have a choice of exchanges. That steering decision is where a conversational coach earns its keep — and where an engine’s +1.2 eval can still leave you in a known draw if you trade the wrong piece.

Tiantian’s classic-manual archive and 残局闯关 mode help if you already read Chinese and want volume. They still assume you will infer the classification system from examples. Ask a coach to name the class before you solve:

"This is chariot and horse versus full guard. Don't give the win yet. Tell me the theoretical result, the first defensive idea, and only then a candidate move."

How Do You Combine an AI Coach With Online Play and Engine Analysis?

You get the most from an AI Xiangqi coach by using it for diagnosis and principles, a play site for volume, and Pikafish for verification — in that order, not the reverse. Starting with the engine produces move-copying; starting with unreviewed blitz produces the same mistakes at higher speed.

A workable weekly loop:

One coaching diagnosis. Open Jenova’s Chinese Chess Coach and state level, time control, and a leak in one message:

"I'm a chess 1600 who learned Xiangqi three months ago. I lose to 卧槽马 and I don't understand when to trade cannons. Treat me as beginner-intermediate. No opening encyclopedia dump."

Ten to twenty rated or bot games on Xiangqi.com or Xiangqi One, same time control. Prefer the platform whose language and piece set you actually see clearly. Desktop still helps for board size; mobile wins for consistency.
One post-mortem, not ten. Paste WXF, Chinese notation, or FEN into the coach and ask for turning points, not a 40-move principal variation. Then run the same game through Xiangqi One’s Pikafish view or Xiangqi.com’s analysis and compare: where did the engine disagree with the principle you were taught?
One classified endgame. Ask for a theoretical-win drill against full guard, or a drawing technique with lone chariot. Stop when accuracy drops; that is the next session’s starting point.
Notation matching. If your club uses 炮八平五, ask in that language. If your server uses C8.5, stay in WXF. Switching notation every week adds friction without adding strength.

For Tiantian users, keep photo-recognition of printed scores and the 60,000-problem bank as homework, then bring why you failed back to a conversational coach. For Pikafish users installing a local engine, follow a UCI setup guide only after you can already explain your plan in words.

Xiangqi.com’s untimed correspondence games (move timers from one to 14 days) fit players who want coaching between moves. Blitz on Xiangqi One fits players who need tempo instincts. Neither format replaces a review in which you have to predict the reply before seeing it.

What Do Xiangqi Instructors Say About Mixing Engines With Human-Style Instruction?

Instructors who care about transferable strength treat engines as graders and coaches as curriculum designers; swapping those roles is the most common way ambitious players stall. Pikafish is extraordinarily strong at choosing moves. It is silent on which mistakes are conceptual, which are visual, and which are just time trouble.

"The players who improve fastest in Xiangqi are not the ones who match engine first choice most often. They are the ones who can say, before they move, whether they are fighting for a screen, a tempo, or a guard-structure change. If you cannot name the fight, a 40-ply line is trivia."

"We also see a consistent plateau among chess transplants. They learn cannon captures in an afternoon and then bleed games to horse blocks, stalemate losses, and perpetual-check forfeits for months. A coach that puts those false analogies on the table in session one is doing more for their rating than another Central Cannon file."

"Persistent memory is not a luxury feature in this domain. Xiangqi strength is an inventory of patterns you personally miss. If the software forgets that you always leave the river horse blockable, it will keep teaching you 顺炮 theory you already understand. That is entertainment, not coaching."

— Jenova Product Team, AI tutoring design for traditional board-game pedagogy

The practical reading of that view is conservative. Use Tiantian or Xiangqi.com puzzles to install mates. Use Xiangqi One or a local Pikafish build to catch tactical blindness. Use a conversational coach to decide which exchanges preserve a 必胜 ending and which “equal” trades throw it away.

How Should International Chess Players Retrain Their Instincts for Xiangqi?

International chess players should retrain by treating every familiar rule as a suspect analogy until a coach or a lost game proves otherwise. The pieces look related; the physics are not.

The highest-cost transfers:

Stalemate is a loss, not a draw. Endgame technique that “should hold” in chess can be a self-mate in Xiangqi.
Perpetual check is a loss for the checking side under standard Xiangqi rules, with a broader forbidden-move family (perpetual check, perpetual kill threat, perpetual chase) in WXF tournament practice.
Horses do not jump freely. The knee block (蹩马腿) is both a defensive resource and a tactical theme.
Elephants do not slice the board. They block at the midpoint and cannot cross the river, so they are palace furniture, not long-range bishops.
The general does not walk the board. Palace confinement plus flying general makes the 5th file and the bottom rank different from king-and-pawn chess.
There is a jumping-capture piece. Until cannon screens are automatic, every “hanging” piece in the chess sense may be bait.

A chess-to-Xiangqi prompt that wastes less time:

"I am 1800 USCF. I already know the river, palace, and cannon capture. Please correct chess analogies as they come up, start with screen management and stalemate endings, and delay Central Cannon theory until I stop walking into 蹩马腿."

Play sites still help this group — Xiangqi.com’s graphical pieces reduce character-reading load — but they will not narrate the analogy. That narration is the coach’s job. Once the false mappings are gone, the same player’s chess-trained calculation often becomes an advantage in sharp 顺炮 (Same Direction Cannons) games, where one tempo decides the result.

The end state is not “find one app.” It is a small stack: Jenova’s Chinese Chess Coach for principles and post-mortems, Xiangqi.com or Xiangqi One for international play and Pikafish checks, and Tiantian Xiangqi if you can use the Chinese-language firehose. That division of labor matches how Xiangqi is actually learned — as literature, as tactics, and as a living tournament game under WXF-shaped rules.