What is your rating worth?
Lichess ratings converted between categories and time controls, from players active in both at the same time, and how busy each queue is
Average — rating of players rated — in — who were playing both at the time.
- 95% interval of the average
- —–—
- Middle half of players
- —–—
- Middle 80% of players
- —–—
- Same-percentile rating (—)
- — (percentile —)
- Player-months behind this figure
- —
- Smoothing window
- —
Rough figure: few players hold both ratings at this level, so the 95% interval is wider than ±50 points.
Too few players held both ratings near this level in this month for a reliable figure. Try a rating nearer the middle of the curve, or another month.
The whole conversion curve
Average — rating at each — rating in —, with the spread of individual players around it. The curve is smoothed across neighbouring ratings and months, more widely where few players hold both ratings, and stops where it would be less precise than about ±100 points. The line is dashed where the average is rough, with a 95% interval wider than ±50. Click the chart to pick a rating. The average regresses toward the middle, so converting back the other way does not return the rating you started with; the same-percentile line is the symmetric alternative.
- Average
- Middle half of players
- Middle 80%
- Rough average (interval wider than ±50)
- Same percentile
- Equal ratings
Drift at —, month by month
Average — rating of — — players each month, with its 95% interval. The pools drift as players join and leave, so a conversion is only good for its own time. Click to pick a month.
Does the method change the answer?
The same conversion at — for the whole of — under other filtering and weighting choices, each with its 95% interval.
Lookup table for —, —
Average rating in each other category for players at each — rating. Values in grey italics are rough (95% interval wider than ±50 points); blank where it would be wider than ±100. Click a row to select that rating.
| Rating | UltraBullet | Bullet | Blitz | Rapid | Classical | Correspondence | Classical (pre-Nov 2017) |
|---|
If every — time control had its own rating, a — player rated — would be rated about:
Higher means that queue is easier: at the same rating, the players you meet there are weaker. These figures are for someone who plays both about equally. If you mostly play one time control you will do better in it than this suggests, and worse in the others.
The gap depends on your rating
How many points higher (above the line) or lower (below it) a — player would be rated in each other time control, at each — rating, in —. Shaded bands are 95% intervals; a line stops where its interval would be wider than ±16 points. Click to pick a rating.
Month by month
The same gaps each month, for — players rated —. Click to pick a month.
How this is measured
- All time controls in a category share one rating, so there is no second rating to compare. Instead the report takes players who played both time controls in the same month.
- In each time control, it compares how they actually scored with what their own and their opponents' ratings predicted, and turns the difference into rating points.
- People do better in whichever time control they play most. To cancel that out, it averages the players who mostly play one with those who mostly play the other. What is left is how much tougher one queue is than the other at the same rating.
- Each figure is smoothed across nearby ratings and months the same way as the conversions between categories: a weighted local fit that widens only where few players play both.
Games per day in each — time control
Average rated lobby games started per day, by month: a stand-in for how many people are in each queue. "other" is every remaining time control in the category.
When the queues are busy, —
Games started in each hour of the day, US Pacific time. Each row is shaded from its own quietest hour (lightest) to its busiest (darkest, with a dot) in five steps, so you can see when each queue fills up whatever its size. The right-hand column is busiest hour ÷ quietest hour: the bigger it is, the more your wait depends on when you play.
- Quietest
- Busiest hour of that time control
Games near your rating, by hour
Games per hour in — where the two players averaged —. Busiest at — Pacific time with — games an hour; quietest at — Pacific time with —.
The data
Every rated lobby game (the "Rated Bullet game", "Rated Blitz game" and similar events) in the Lichess standard-chess open database from — to —: — games once bot games are removed. Tournament games are not in this copy of the data, which suits the queue-size measure. The 2026 figures cover January to September only.
Lichess places a game in a category by its estimated duration, base time plus 40 times the increment: under 30 seconds is UltraBullet (here only ¼+0), under 3 minutes Bullet, under 8 Blitz, under 25 Rapid, and the rest Classical. Each category has one Glicko-2 rating shared by all its time controls. Rapid did not exist before November 2017, when Classical covered everything from 8 minutes up, so the old Classical pool is shown on its own as "Classical (pre-Nov 2017)". New Rapid ratings were seeded from Classical at the split.
Between categories
- Only contemporaneous play counts. A pair of ratings is recorded at a moment when both were current: the player had just played in one category and had played in the other within the last 30 days.
- No provisional ratings. Rating deviation is not in the data, so it is worked out from each game's rating change. Both ratings must be established (deviation 75 or less, where Lichess shows a rating as rankable) and the player must have at least 20 games in each category.
- Frequent dual players count most. Each player-month gets a weight equal to the smaller of their game counts in the two categories that month, capped at 30, so someone who plays both every day counts far more than someone who dabbles, without letting a handful of accounts dominate.
- Drift is handled month by month. Every figure belongs to one month and borrows only from the months around it, so a conversion always describes the pools as they were at that time.
- Two kinds of answer. The average is the expected rating in the other category for people at your rating. Because nobody's two ratings move in lockstep, it regresses toward the middle, so A to B and B to A are not inverses. The same-percentile rating matches ranks among the dual players instead, and is symmetric.
- Smoothing. Each figure is a weighted fit through the player-months nearby in both rating and time: a local-linear smoother (the core of loess) with weights falling off smoothly with distance in rating points and in months, and a linear trend in time so the latest months are not dragged toward the past. The window is ±25 rating points and ±2.5 months where data is dense, and widens step by step to at most ±150 points and ±13 months until about 1,500 player-months are inside it, so busy pairs keep near-monthly detail and rare ones borrow from their neighbours. The spread of players comes from the same window after moving each neighbour along the fitted slope. The same-percentile rating and the method comparison are yearly.
- Always rising. A higher rating in one category should never mean a lower expected rating in the other, but where few players hold both, noise can make a smoothed curve dip. Each month's curve is therefore made non-decreasing (a weighted isotonic regression): a stretch that dips becomes a flat run at its precision-weighted average, and rising stretches are left alone. The spread of players is treated the same way. This moved about 3% of the figures, mostly rough ones, and almost always by less than their 95% interval.
- Intervals. 95% intervals come from a jackknife over 16 groups of players. The middle half and middle 80% describe how individual players spread around the average.
Within a category
- Time controls in one category move the same rating, so there are no two ratings to compare. Instead, for each player who played two of them in the same month (at least 5 games in each), the report measures how much better or worse they scored in each than both players' ratings predicted, and takes the difference.
- Player-months are combined with random-effects weights, so players with many games count more but saturate. A player's rating here is their average rating in those games.
- Each month is labelled by which time control had at least two thirds of the player's games in that pair. Players do better in the one they play most, so the figure shown is the midpoint between those who mostly play one and those who mostly play the other: the gap for someone with no preference. Players who split their games evenly land close to that midpoint, which supports it.
- Each group's average is a weighted local fit in rating and month, like the conversions between categories: ±25 rating points and ±2.5 months where data is dense, widening (to at most ±150 points and ±13 months) until each group has about 400 player-months. A figure is shown only where its 95% interval is narrower than ±16 points.
- The time controls compared are the current home-page quick pairings plus each category's three most played over the whole data, which adds only ½+0 Bullet and 15+15 Classical.
Limits
- Dual players are not a random sample of everyone. A 1500 Rapid player who also plays Blitz may differ from one who never does.
- Lichess changed how fast rating deviation grows between games in August to September 2019 and again in January 2020, which shifts how quickly ratings settle around those dates.
- Near the ends of each curve, and for the rarer categories, few players hold both ratings. The report marks a conversion as rough when its 95% interval is wider than ±50 points and leaves it out when wider than ±100. Read a rough figure as a range.