What real ELO would you assign to the chess.com bots?

E.g. Antonio "1500", Isabel "1600" etc. The free ones. 

What would you guess their actual ELO would be? Or is it just not possible because of how they operate, with the sorts of blunders which almost no one would ever make, to try to "balance" their play?

I feel they are more overrated than that. I feel like Antonio "1500" is 900-1000 at most, and Isabel "1600" is maybe 1250 tops. But like you said maybe there's no way to really compare to real ELO. The mistakes they make are very artificial. 

As someone who has thought a lot about making chess engines (but still too lazy to progress far with making one) I feel like it would be easy to do a lot better than these bots, in terms of making a lower-rated player. Just pick the right level of move depth, have an inconsistent move depth and have it fall victim to certain traps and blind spots. Like tunnel-vision ignoring bishops, forgetting to protect the rooks from queen attacks etc. Just seems like these bots were made very lazily.

What you shouldn't do is have these random blunders which are nothing like humans. Where they literally just moved a knight or bishop, you attack it, and then they completely ignore that it was attacked. That's like the opposite of tunnel vision.

Actually when you finish a game you get the report with approximate ratings of both players. Speaking about Antonio his rating was 1100 all the time I played with him.

Every beginner bot: 100-200

Every intermediate bot: 500-600

Every advanced bot: 800-1000

Every master bot: 1900-2500

KieferSmith are you talking about the USCF ratings? I can beat Antonio and my USCF rating is only 637.

I beat Antonio, Isabel and Wally and they are nothing like their elo rating

Antonio = 1100

Isabel = 1250

Wally = 1500

I'm only 500 elo

Depends what time control you mean. The bots always play at the same speed no matter how slow the player plays, so their real classical and rapid ratings would be significantly lower than their blitz rating, which is the speed they actually play at.

Very true

How do you add time control, btw?

There's no feature for that when playing bots, but if you wanted to give the bots a "real" elo, step 1 would be to decide what time control. Probably something where they won't accidentally run out of time or leave too much time unused, as they have zero clock management logic.

Antonio probably plays like a 1000 rated human player on chess.com

I would say because of the very odd nature of the bots, obviously not being able to play like lower-rated humans, just half their elo and subtract 100, until about 1600 (above that usually the rating gap closes a lot). I recently noticed that Antonio was moved into the locked section instead of the free column, but I beat him first when I was about 700 elo and he was blundering a lot, if I recall correctly he is rated 1500 so about 650 would be his rating. I think Martin can use this same logic which means he would be -50 elo

Edit: I'm ~850 elo now and I just beat Isabel (1600), so I think 700 is pretty much accurate for her

Edit 2: I also beat Wally (1800) losing once to a careless blunder so 1-1 and it was pretty even so yeah 800-900 elo for him

As long as bots aren't playing with AI they will always be bad, it's completely meaningless for the bot to know the best move then INTENTIONALLY blunder a queen. Playing with them while having a proper internet connection is a waste of time.

fr bro im 500 i start playing chess a week ago and i find it bit sus how i can beat these 1100 elo and its to good to be true and recently i played lvl 3 stockfish and after like 4-5 games i finally got to beat it can someone tell me the real elo of these bots ??

That's playing with humans is waste of time. Bots don't just randomly blunder queens, that's not true. Their play is now humanized very well. And their rating is precise and corresponds to actual FIDE ratings. Chess.com players are underrated, that's for sure.

@grizzlygizmoo if you had to play 5 games to finally beat a bot, your level is 241 Elo lower than level of that bot, not equal or higher. Provided that you've used 5min +5sec per move increment control for yourself (if your game was longer, then your actual Elo is even lower)

"Can usually beat" isn't specific enough to justify that specific reduction you've suggested. Chess.com uses Komodo, it's calibrated to 5|5 Blitz or 10|0 Rapid. But did you use any time control when playing with bots? This matters a lot. Quote:

If the human is treating the games like serious tournament games (say 90 min + 30 sec) then the corresponding elo values would be a few hundred lower.

Next, how many attempts and how many wins? Did you count losses (due to blunder, distraction - doesn't matter) or pretend they didn't happen? This matters a lot too. Because, if you're winning only 2 out of 10 games fairly against any bot you are 241 Elo weaker than that bot. Here's a table for rating difference calculation (0.5 point in score can come from draws):

Next, bots are calibrated in Engine tournaments where FIDE-rated GMs also competed. Therefore ratings of bots are FIDE-like. However, chess.com ratings are far from FIDE. So there's no point in direct comparison.