Skip to content

Guess 2/3 of the Average

Pick a number. The winner is whoever comes closest to 2/3 of everyone's average this week. Free, no sign-up. We save guesses, with your IP address and country, to run the rounds.

By the Helibrain teamReviewed by Robert, biophysics PhD candidatePublished Last reviewed

Loading this week's round…

How to read your result

  • Your level shows how many steps ahead you thought. A guess near 50 is level 0: no thought about the others. Near 33 is level 1: "people will guess about 50". Near 22 is level 2: "people will think one step ahead". Each step is 2/3 of the one before [1].
  • Game theory's answer is 0. If everyone reasoned all the way, every guess would shrink to 0. Real players never do this, so 0 has never won a public contest [1][2].
  • Can never win: a guess above 66.7 can't win, because even if everyone guessed 100, the target would be 66.7.
  • This week's winner appears when the round ends (each round is a week, Monday to Sunday, UTC). We never show the live average before that: it would spoil the game. A round needs at least 30 players to have a result.

What is the best number to guess?

There is no single best number: it depends on what everyone else guesses. In big public contests, the target landed between about 13 and 22. Financial Times readers averaged 18.9, so the target was 12.6 [2]. In a Danish newspaper contest with 19,196 readers, the average was 32.4 and the target 21.6 [3].

Bar chart: average guess and target in four published contests. Lab study 1995: average 36.7, target 24.5. Politiken 2005: average 32.4, target 21.6. Spektrum 1997: average 22.1, target 14.7. Financial Times 1997: average 18.9, target 12.6.
Figure 1. The average guess and the 2/3 target in four published contests. Data: Nagel 1995, first round, 67 students [1]; Politiken 2005, 19,196 readers [3]; Spektrum der Wissenschaft 1997, 2,728 readers, and Financial Times 1997, 1,476 readers [2].

Quick facts

  • What it shows: how many steps ahead you reason about what other people will do. Economists call this level-k thinking [1][4].
  • Classic experiment: in the first round of Rosemarie Nagel's 1995 study, students guessed 36.7 on average, and the most common guesses sat near 33 and 22 [1].
  • Big contests: 1,476 Financial Times readers averaged 18.9 (the winning guess was 13), and 19,196 Politiken readers averaged 32.4 [2][3].
  • The pull of 33: in a 2015 New York Times version with about 56,000 readers, 33 was the most popular guess [5].
  • Where the idea comes from: the economist John Maynard Keynes compared the stock market to a beauty contest where you win by picking the face others will pick [6].

How the game works

  1. Pick a whole number from 0 to 100 and lock it in.
  2. Everyone's guesses this week are averaged when the round closes, and the target is 2/3 of that average.
  3. The closest guess wins. Ties share the win. On Monday, the page shows last week's players, average, target, winning guess and how every number was guessed.

One guess per person counts in each round. We count one guess per browser and one per internet connection, so people sharing a network (a school, an office, a family) count once a week. A second guess is saved but doesn't count. Right after you guess, the page shows where your number sits on the reasoning ladder and how it would have done in the famous contests above.

What the game measures, and where it comes from

Economists call this game the p-beauty contest, with p = 2/3 here. Rosemarie Nagel made it a famous lab experiment in 1995 [1]. Her finding: people don't reason all the way to 0, and they don't guess at random either. Most stop after one or two steps of "they think that I think".

That idea, that people reason a limited number of steps about others, became the level-k and cognitive hierarchy models of strategic thinking [4][7]. The newspaper contests that followed, with thousands of readers each, showed the same steps at 33 and 22, plus a group who chose 0 or 1 [2].

What this game can and can't tell you

Your level describes one guess in one game. It isn't a measure of intelligence, and a deeper level isn't better: the winner is whoever predicts the crowd best, and the crowd usually stops at one or two steps. A level-4 guess is clever reasoning that usually ends up below the target.

The players here are whoever visits that week, not a fixed sample, so the target moves from week to week. The famous contests used slightly different rules: the Financial Times allowed whole numbers only, and Politiken and Spektrum allowed decimals [2][3].

Frequently asked questions

What is the answer to guess 2/3 of the average?

Guess 2/3 of the average has no fixed answer. Game theory says 0, because every player has a reason to go lower than everyone else. But real people stop after one or two steps, so in published contests the winning number was between 13 and 21.6 [2][3]. The best guess is 2/3 of what you expect others to guess on average.

Why doesn't 0 win?

Zero would only win if almost everyone reasoned all the way down. In real contests, many people guess around 33 or 22, and some guess high numbers, which pulls the average up. In the Financial Times contest, the average was 18.9 and the target 12.6, far from 0 [2].

What does "level 2 thinker" mean?

A level-2 thinker expects other people to think one step ahead (they will guess about 33) and goes one step further: 2/3 of 33 is about 22. In the Helibrain game, your level is the step on the ladder (50, 33, 22, 15, 10) that your guess is closest to [1].

Can I play more than once?

Yes, but only your first guess each week counts in the round. Later guesses are saved and shown on the ladder, but they don't count, so nobody can tip the result by guessing many times. A new round starts every Monday.

Is the Helibrain 2/3 game really free?

Yes. The Helibrain 2/3 game and every weekly result are free: no sign-up, no email, no payment. We save each guess with your IP address and country to run the rounds fairly, as our privacy policy explains. Helibrain makes money from its app, which has a free plan and paid plans; you don't need the app to play.

See all 37 free brain tests

Want to work on this daily? The Helibrain app has a free plan and paid plans.

Sources

  1. Nagel R (1995). Unraveling in guessing games: an experimental study. American Economic Review 85(5):1313–1326. www.jstor.org/stable/2950991
  2. Bosch-Domènech A, Montalvo JG, Nagel R, Satorra A (2002). One, two, (three), infinity, …: newspaper and lab beauty-contest experiments. American Economic Review 92(5):1687–1701. doi.org/10.1257/000282802762024737 (numbers from the 1999 working paper: econ-papers.upf.edu/papers/438.pdf)
  3. Schou A (2005). Gæt-et-tal konkurrence afslører at vi er irrationelle. Politiken, 22 September 2005. politiken.dk/oekonomi/art5698526
  4. Camerer CF, Ho T-H, Chong J-K (2004). A cognitive hierarchy model of games. Quarterly Journal of Economics 119(3):861–898. doi.org/10.1162/0033553041502225
  5. Quealy K (2015). How readers fared in Upshot's number puzzle. The New York Times, 20 August 2015. www.nytimes.com/2015/08/21/upshot/how-readers-fared-in-upshots-number-puzzle.html
  6. Keynes JM (1936). The General Theory of Employment, Interest and Money, chapter 12. London: Macmillan.
  7. Stahl DO, Wilson PW (1995). On players' models of other players: theory and experimental evidence. Games and Economic Behavior 10(1):218–254. doi.org/10.1006/game.1995.1031

How we made this page: written by the Helibrain team from the studies listed above, and reviewed by Robert (biophysics PhD candidate) on 4 November 2026. We update it when the sources change.