How realistic are the bot ratings?

Hi Community!

I won against 1600 bots and usually win against 1100 or 1200 bots. But against real people it's hard to win against 900-1000 rated players and I usually lose against 1000+ rated players. So could it be, that these bot rating numbers are not very accurate?

Does anyone know how these bot ratings are calculated?

Greetings from Germany (I really like playing against players all around the globe, thanks chess.com ;-))

The ratings of the bots are too high, as you've noticed.

I assume this is because chess.com wants to attract new players, not discourage them... so new players get used to chess.com bots and only later become discouraged by playing humans. That's their basic plan.

The bots play at about the level of a human at a time control of ten minutes. To illustrate: I'm rated about 1200; in ten-minute games I beat the 1200 bot about half the time; in thirty-minute games I'm about even with a 1400 bot; in five-minute games I'm barely up there with a 1000 bot.

Bots might play positions or make moves that human players would never consider, either because they lack a certain type of human intuition or because they are programmed to reach certain positions. This can create unrealistic scenarios that don't often occur in real human games.

bruh

My rating is 250-275 and i beat a 1000 bot

:)

I think we know that I once beat a 2000 rated bot and the performance rate of the bot was like 1200-1500.

I personally think the correct answer is "Not that Realistic depending how good you played"

If you played bad, they act like 2000 rated but if you play good, they don't act like a 2000 and sometimes its vice versa!

Truth

It doesn't matter because you just feel how fake it is, like what bludering a rook and when next 5 moves playing 5 move tactic. BS

Bots can have far more sophisticated logic and settings than just performing deep searches and occasionally making random blunders. If you encounter a poorly designed or badly configured bot, it might behave that way, but many modern bots are much more advanced. For example, Komodo has customizable settings (as shown in the screenshot), which allow it to simulate different playing styles and skill levels. Komodo is also used on Chess.com, but how each bot is configured on the platform - we don't know. Maybe some are badly configured. However, there’s no setting for "random blunders", that's just a myth. See:

When it comes to measuring a bot’s Elo rating, this is typically done through engine tournaments. You create a pool of bots and have them play each other in thousands of games. Based on their win/loss records, you can calculate their Elo ratings with high precision.

In fact, it’s easier to measure a bot’s Elo compared to a human’s, because bots can play numerous games without fatigue, allowing for more accurate statistical analysis. So, you can trust the Elo ratings assigned to bots - those ratings are reliable. But it's a mistake to compare them to rating of chess.com players. And here's why: the Elo ratings of modern chess engines tie back to the early 2000s when engines were compared against Grandmasters with FIDE ratings of 2700-2800. As newer engines emerged, they were rated by competing against older engines, forming a continuous chain. So, today’s bot ratings are still linked to FIDE ratings from the early 2000s. It might be tempting to adjust the ratings of bots to better match player ratings on Chess.com, but doing so would break their connection to the global / historical pool of bots and would require constant recalculations, as the player pool changes frequently. Rating of players on chess.com is a different story, do not even try comparing to rating of bots.

It makes more sense to estimate your "FIDE-2000 rating" by playing against bots, because that rating is the most stable and accurate and that essentially places you in the same pool as players like Kasparov, offering a better sense of where you stand in your progress.