This Elo rating calculator runs the world’s most borrowed rating system: expected score from the rating gap, E = 1/(1 + 10^(−Δ/400)), and the post-game update R’ = R + K × (actual − expected) for wins, losses and draws — with your choice of K-factor.

–

The Logistic Heart of Competitive Ratings

Arpad Elo’s insight (chess, 1960s) was that competitive strength differences translate to win probabilities through a logistic curve in base 10: a 100-point edge means the stronger player expects 0.640 of the points — exactly the defaults’ setup (1600 vs 1500), where a win moves A to 1612 (+11.5: winning was likely, so it teaches little) while a loss would crash A to 1580 (−20.5: the upset is information). That asymmetry — cheap expected wins, expensive shocking losses — is the entire philosophy in one line: ratings are beliefs, and results revise them in proportion to surprise. FIDE adopted Elo for chess in 1970; the system then colonized football rankings, tennis, esports ladders, video-game matchmaking and NFL power models, always with local dialects (different K, different divisor, draws handled at 0.5, team versions with home advantage added to the gap before the curve). This calculator implements the canonical form with adjustable K so any dialect’s arithmetic can be reproduced.

The K-factor is the system’s thermostat: it caps how much one game can teach. Chess federations taper it (40 for new players whose ratings are guesses, 20 standard, 10 for masters whose ratings are settled), sportsbooks’ models effectively tune it to minimize prediction error, and online ladders crank it during placement matches. Choose yours consciously: the update result here shows both players’ new ratings so the transfer symmetry (zero-sum at equal K) stays visible — break K symmetry between players (provisional vs established) and the system quietly creates or destroys points, a feature leagues exploit and purists debate.

Worked Examples (Verified)

  • Defaults (1600 vs 1500, K = 32, A wins): expected 64.0% → A 1612, B 1488 (±11.5 transferred).
  • The draw: same players draw → A = 1600 + 32(0.5 − 0.640) = 1596, B = 1504: the favorite pays 4.5 points for failing to win, the underdog collects them.
  • The upset: B wins → A drops to 1600 − 20.5 = 1580, B jumps to 1520: the biggest swing K allows at this gap short of the full expected reversal.

Elo Discipline

  • Ratings are pool-relative: a 1600 in one federation/game/era is not a 1600 in another — the numbers only speak within their ecosystem and era (inflation is real).
  • Match the K to the implementation you’re verifying: FIDE, Chess.com (Glicko variant), FIFA (SUM system post-2018, Elo-derived) and FiveThirtyEight’s NFL Elo each differ in details — starting ratings, home edge (+~48–65 pts in many team versions), and margin-of-victory multipliers.
  • Expected score is a long-run frequency claim: 64% doesn’t mean “wins this game 64% of the time” in any mystical sense — it calibrates over hundreds of such matchups.
  • Sequential updates are order-sensitive in batches: tournaments update game-by-game or round-by-round by rule — replicate the host’s order, not your own.
  • New players need provisional handling (higher K, placement games): feeding a beginner’s first games at master K destabilizes both ratings.

Related Tools

Frequently Asked Questions

How do you calculate Elo ratings?

Expected score E = 1/(1 + 10^(−(RA − RB)/400)); new rating R’ = R + K × (score − E), with score 1/0.5/0 for win/draw/loss. Defaults: 1600 vs 1500 → E = 0.640; A wins at K = 32 → 1600 + 32 × 0.360 = 1612. This calculator returns expected percentages and both updated ratings for any K and result.

What does a 100-point Elo difference mean?

The higher-rated player is expected to score 64% of the points (over many games, counting draws as halves). 200 points ≈ 76%, 400 ≈ 91%, and 0 points is exactly 50%. The curve is logistic, so equal gaps at different heights mean the same expectation — a 2400 vs 2300 pairing mirrors 1600 vs 1500 probabilistically, though the higher pair draws far more often in chess practice.

What K-factor should I use?

The one your competition uses: FIDE chess — 40 (first 30 games/new), 20 (established below 2400), 10 (2400+); USCF uses a graduated table; casual ladders often 32; team analytics models typically 20–32 per game with sport-specific tweaks. K is a policy choice (stability vs responsiveness), not a physical constant — the calculator exposes it rather than hiding a default philosophy.

Why does the favorite gain fewer points for winning?

Because the update is (actual − expected): beating a weaker player confirms the rating (small surprise, small change), while losing to one contradicts it (large surprise, large change). This makes ratings self-correcting measurements rather than win counters — a 1600 who farms 1200s gains nearly nothing per win and is one upset away from donating a pile, which is exactly the incentive structure honest ladders want.

How are draws handled?

As half a win for each player in the score term: the favorite’s rating falls on a draw (scored below expectation) and the underdog’s rises. In the defaults’ pairing, a draw moves A 1600 → 1596 and B 1500 → 1504. Sports without draws skip the case; competitions that replay draws (tennis can’t draw) never invoke it.

Elo vs Glicko vs TrueSkill?

Glicko adds a ratings-deviation (uncertainty) term — inactive players’ ratings become less certain and move faster on return; TrueSkill (Microsoft) extends the idea to teams and multiplayer via Bayesian skill distributions. All three share Elo’s logistic/expectation skeleton. This page implements the original Elo arithmetic — the baseline every variant still quotes.

Can Elo rate teams?

Yes — treat the team as the player, add a home-advantage bump (commonly ~48–65 rating points) to the home side’s effective rating, and consider margin-of-victory multipliers (FiveThirtyEight’s NFL model used them). Season rollovers often regress ratings toward the mean (e.g., ⅓ back to 1505) to model roster churn — league-specific layers atop the same core this calculator runs.

Is this Elo rating calculator free?

Yes — free, instant, no sign-up. Enter two ratings, a K-factor and the result for expected scores and both updated ratings.