Chess Win Probability Calculator

Where Elo ratings come from

Arpad Elo, a Hungarian-American physics professor, invented the rating system in the 1950s to replace the older (and statistically unsound) Harkness system used by the US Chess Federation. The Elo system was adopted by USCF in 1960 and FIDE (international chess body) in 1970. It’s now used in chess, but also Scrabble, Go, esports, online dating apps, and even Facebook’s old “Hot or Not” rankings.

The core idea: each player has a rating that represents their playing strength. When two players meet, the difference in their ratings predicts the expected outcome.

The expected score formula

Expected Score (for the higher-rated player) = 1 ÷ (1 + 10^((opponent − you) ÷ 400))

The “400” is calibrated so that a 200-point gap produces a roughly 76%/24% expected split (treating draws as half-wins for each side).

Expected score ranges from 0 to 1:

• 1.0 = certain win
• 0.5 = certain draw
• 0.0 = certain loss
• 0.76 = expected to win 3 of every 4 games on average

Rating differences and expected outcomes

The draw column comes from fitting a curve to classical tournament results, not from Elo itself. Elo gives you the expected score and says nothing at all about how it splits between wins and draws, which is why every site quotes a slightly different W/D/L breakdown for the same gap. The shape everyone agrees on: draws peak between equals, around 30% in classical play, and decay roughly as a Gaussian in the rating gap. They never reach zero. Even at 500 points apart, one game in sixteen is drawn.

Time control moves this a lot. Blitz and bullet produce far fewer draws at every gap, because there is not enough time to convert a small edge into a technical draw and blunders decide games instead.

Rating ranges in chess

The world average rating is around 1200-1400 for active tournament players. Online platforms (Chess.com, Lichess) use different scales, typically 100-200 points higher than the equivalent USCF/FIDE rating because of the player pool composition.

Rating change formula

When you play a rated game, your rating changes based on actual vs expected outcome:

New rating = Old rating + K × (Actual − Expected)

Where K is the rating change factor:

• K = 40 for new players (under 30 games)
• K = 20 for established players under 2400
• K = 10 for players 2400+
• K = 10 for FIDE titled players

Example: 1600-rated player beats 1800-rated opponent

• Expected score: 0.24
• Actual: 1.0
• Change: K × (1.0 − 0.24) = 20 × 0.76 = +15.2 rating points
• Opponent loses the same amount

Drawing the 1800 gives the 1600 player a 0.50 actual − 0.24 expected = +0.26 × 20 = +5.2 points.

Conversely, the 1800 expecting to win and only drawing loses 5.2 points.

The Glicko system, the modern alternative

Mark Glickman developed Glicko (now Glicko-2) as an improvement on Elo:

• Tracks rating uncertainty (RD, or rating deviation) separately
• Players with few recent games have high uncertainty; their rating moves more
• Active players have low uncertainty; their rating moves less per game
• More accurate during rapid improvement periods

Chess.com, Lichess, and most modern online chess use Glicko-2. FIDE still uses traditional Elo with K-factor adjustments. The two systems converge for established players but Glicko handles new and returning players better.

FIDE vs USCF vs online ratings

The biggest source of confusion in chess ratings: each platform has its own scale.

So a USCF 1500 player would be roughly Chess.com 1700 rapid, Lichess 1600 classical. The differences come from:

• Player pool composition
• Time control effects (faster = more error = higher variance)
• Initial ratings (Lichess starts everyone at 1500; Chess.com starts at 1200; USCF tournaments seed based on initial tournaments)

Time control affects rating

Strong players are not equally strong at all time controls:

A player rated 2000 classical might be 1800 rapid, 1700 blitz, 1600 bullet. Or they might be the reverse. Some players excel at fast play, others at slow.

The Magnus Carlsen factor

Carlsen’s peak FIDE rating of 2882 (2014) is the highest ever recorded. Run it through the formula:

• vs a 2800 player, an 82-point gap: expected score 0.616
• vs 2700, where most super-GMs sit: 0.740
• vs 2500, a typical grandmaster: 0.900
• vs 1500, a decent club player: 0.9996

That last one is worth sitting with. Even against the best player who has ever been measured, a 1500 scores four points in ten thousand games, and most of those are draws rather than wins.

The system does get shaky at the very top, because there are not enough 2800-rated opponents to calibrate a 2882 against. But the ordering holds, and so does the rough shape.

Rating inflation/deflation debate

Chess fans regularly debate whether modern ratings are inflated compared to past eras. The arguments:

Pro-inflation: Modern computing makes openings sharper; players use engine prep; ratings of 2700+ are more common than in the 1990s.

Anti-inflation: Modern players have better training tools; skill has genuinely improved. Comparing Kasparov (1990s peak 2851) to Carlsen (peak 2882) shows real improvement.

Most statisticians conclude there’s no significant inflation, because the system is mathematically self-correcting. Strong players replace older strong players over time.

Performance rating

A different concept from regular rating: performance rating is the rating a player would need to achieve their tournament result.

Performance = Average opponent rating + 800 × (score / games − 0.5)

If you score 4/5 against an average opponent rated 1800: performance = 1800 + 800 × (0.8 − 0.5) = 1800 + 240 = 2040.

A 200-point performance over your rating in a serious tournament is a strong result. Tournament norms (IM, GM) require multiple tournaments with specific performance levels.

The one factor nobody puts in the formula

Elo says nothing about who has White. In classical chess between equally rated players, White scores about 54%, which is worth roughly 35 rating points. That is more than the gap between two adjacent rating classes at the club level, decided before a single move is played.

The advantage shrinks as you go down the rating ladder (below about 1600 it is closer to 52%) and it has been slowly shrinking at the top for decades as opening theory has caught up with the first-move initiative. The colour selector above applies the 35-point adjustment so you can see what it does to your own numbers. In a two-game match with colours reversed it cancels out exactly, which is precisely why matches are played that way.