I do realise that a similar question might have been asked before; however, all intel about the topic that I have managed to gather to date seems to be outdated: almost everywhere I search, it is said that PlayMagnus age 10's estimated elo is 1500-1700, but I believe him to be much stronger than that. My Rapid elo on chess.com is 1850, but in order to beat him, I need to concentrate intensely, and even then I'm more likely to draw or lose than win. There is no doubt in my mind that the bot is stronger than me - I would estimate by about 200-300 elo points.

To this end, here is the question: am I simply significantly overrated, or is age 10 Magnus' true elo actually higher than what is commonly cited?


It's important to note the PlayMagnus app is running the Glaurung chess engine. The Glaurung chess engine is the forerunner of Stockfish. It's roughly about the same level as Shredder/Junior (link), both former winners in the computer chess championships. Very far away from the modern Stockfish but also far stronger than your chess.com's rating. You won't stand a chance unless the engine artificially plays bad chess. Unfortunately, there is no good way to make an engine play weaker while human-style chess. Lot's of different techniques, but nothing is satisfactory.

However, the PlayMagnus group has professional chess programmers working full-time to tune the engine for each age group. While I don't know their internal benchmark, the fact that they're pushing the app out means they're satisfied with the performance. Magnus Carlsen at age 10 was about ELO 2000 (from Wikipedia), so that'd be the target ELO. A tuned-down engine will not be playing consistently at human-level ELO 2000, but it should be closed given after so many years of professional development.

A very roughly ELO 2000+ (computer engines tend to biased up) is what you should expect. That's consistent with your observations.

  • Yeah, this is probably a better answer than mine. Mar 15 at 15:52
  • 1
    "Unfortunately, there is no good way to make an engine play weaker while human-style chess. Lot's of different techniques, but nothing is satisfactory." This couldn't be more accurate. The highly artificial means through which the target elo is lowered are painfully obvious. In the opening and early middlegame phases, the bot plays like a regular 1600-1700, which usually allows to me to gain an advantage over it early on, but as soon as the advantage gets too big, the bot turns superhuman mode and starts playing like a 2200+, up until the endgame, at which point it drops to <1500.
    – Max
    Mar 15 at 22:48
  • It seems like the devs were attempting to reflect Magnus age 10's actual strengths and weaknesses by breaking the game down into 3 stages and making the bot play at variable elo throughout the 3, but the way they ended up implementing it made the bot even less human-like and its playstyle even less realistic.
    – Max
    Mar 15 at 22:55
  • @Max You could be right, but I don't know. The exact algorithm being used is not available to the public.
    – SmallChess
    Mar 15 at 22:57

The magnus 10 bot is so horribly coded I am not even joking. I have been playing chess for 4 years and I was able to beat age 9 within the first year but magnus age 10 I can only beat if I'm incredibly careful and play as if it's a tournament game.

The bot makes almost all perfect moves until it has a +5 advantage after which it plays inaccurately but still not bad enough for you to get a good lead. Sometimes it will have a mate in 3 but it will play crazy moves that do literally nothing but keep your king trapped. This is supposed to simulate a certain elo but it does it horribly.

I'll play a game with it rn and I'll send a link of the game's analysis.

Magnus age 10 bot is at least 2000 rated in terms of chess com elo.

  • 1
    I agree, there is something very much off about the way it's coded. I think the way it works is: 1) In the opening phase, it plays main-line theory 2) in the early middlegame, it's supposed to give you chances, but is still playing moves that are at worst inaccuracies 3) it will keep playing like that until you get a significant enough advantage (around 200-300 centipawns, usually), at which point it will start playing perfect moves 4) it will keep playing perfect moves until the endgame, when it will start making random blunders...
    – Max
    Mar 15 at 23:07
  • 1
    ...or until it gets a completely winning position, in which case it will start playing random moves which prolong the game to simulate human inaccuracies. Either way, playing the bot just doesn't feel right at all. I feel that, while chess.com's lower-rated bots are also awful, its advanced and master bots feel much more human and realistic than PlayMagnus' bots of equivalent strength. It was much easier for me to imagine that I'm playing against a human opponent, rather than a cyborg programmed to occasionally make moves that make no sense, in a game against, e.g. Agadmator's bot.
    – Max
    Mar 15 at 23:08

It's impossible to answer. It doesn't have a true Elo.

There is hardly such a thing as "True Elo", because there are many different ratings and they can't be directly compared. But the most widely recognized, the FIDE rating, is based on rated tournament games. People are focused, have a lot of time, and they play other such people.

The rating you mentioned is your chess.com rapid rating. Chess.com uses a different rating system with different parameters, a different starting ratings, and a different pool of players. And it's for rapid, on an online site, where people usually aren't as focused as they are in a tournament game (and that differs a lot between players). It's not a "True Elo" either and it can't be directly compared to other ratings like FIDE.

Worse, computers, when they are tweaked to play at a lower strength than the default, don't become weaker in the same way that humans do. They are either given an extremely short amount of time, or they are sometimes forced to randomly play a worse move. Their resulting strength may still depend on your phone's CPU speed. It's very hard to say to what kind of human strength the resulting engine would correspond.

In short, if they manage to roughly come within a few hundred points of what they're aiming at, then I would consider that good.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.