I toast to LeelaqueenOdds and its creators. I raise my wine glass high and shout: Long live the queen. We have a queen!!

[Father]

I think that we need some competition with significant money prizes to see what humans can really do against LeelaQueenOdds without Qd1.

My idea is that every player choose the time control that he prefers that has to be of the time X minutes per game + Y seconds per game when X and Y are integers and Y is the closest to X/2 that means the following time controls are allowed
1+0.1+1,2+1,3+1,3+2,4+2,5+2,5+3,6+3,7+3,7+4,8+4,9+4,9+5,10+5

Every player play a match of 10 games against LeelaQueenOdds when the ranking is based on the cube of number of points divided by X+Y

For example losing 6-4 at 1+0 gives 4^3 points that is 64 and winning 10-0 at 10+5 gives 10^3/15 points that is 66+2/3 so winning 10-0 at 10+5 is better than losing 6-4 at 1+0 but worse than losing 5.5-4.5 at 1+0 that gives 91.125

We can give prizes to the first 3 players that are proptional to the number that they get in the ranking.

What is your opinion?

Joel Benjamin (SuperEasy) consistently scores 9, 9.5, or 10 out of 10 at 8'3" (not on your list but about same as 7'4"), so he would get between 65 and 91 points each time based on past results. I think it would be an interesting contest, but are you talking about an open contest on the internet, where cheating would be a concern, or a monitored event by you in Israel? It is unfortunate that any time money prizes are offered to the general public for similar events, cheating seems to be a real problem. It's fairly easy to catch cheaters in bullet chess events, where the cheating is generally automated, but in Rapid it's extremely hard to catch sporadic cheaters. Assuming a monitored event, I think most players will select the shortest time control at which they expect to score like 90% or so, which for most people means the maximum 10'5" will be chosen. If you get GMs then perhaps something like 5'3" will be popular. Results for the bot are pretty amazing, especially in the bullet range. Of course no one can compete at 1'0", but even at 2'0" performance of the bot is astronomical; against players above 2600 LiChess bullet it won 238 games vs. only 4 losses (ten draws). I don't think that incentives matter for bullet games, they are rarely drawn so it hardly matters if players resign in drawable positions, and they surely don't move too fast. But the longer the time limit, the more important incentive and motivation become, so I agree that if we want to measure performance in Rapid, some incentives are needed.

I consider organizing this competition.

I plan to donate 3000 shekels to the prize money that is something near 800 dollars.
I wonder how many players can practically play at the same time(I may invite only limited number of players to the competition because I assume Leela cannot play too many games at the same time and I do not want Leela to perofrm worse because of playing too many games at the same time.

If it is possible it is better not to allow everybody to play against lc0 in the time cf the competition but only the players who play against it.

Good evening Mr. Uri Blass. If at some point they wanted me to be part of the humans in the competition, I would be willing to compete. My horse Catecan and I are. citizens of the world, and I would like to represent a common flag: "That of being human" riding through the roads of the
life. For my part always grateful.

[Father]

I think that we need some competition with significant money prizes to see what humans can really do against LeelaQueenOdds without Qd1.

My idea is that every player choose the time control that he prefers that has to be of the time X minutes per game + Y seconds per game when X and Y are integers and Y is the closest to X/2 that means the following time controls are allowed
1+0.1+1,2+1,3+1,3+2,4+2,5+2,5+3,6+3,7+3,7+4,8+4,9+4,9+5,10+5

Every player play a match of 10 games against LeelaQueenOdds when the ranking is based on the cube of number of points divided by X+Y

For example losing 6-4 at 1+0 gives 4^3 points that is 64 and winning 10-0 at 10+5 gives 10^3/15 points that is 66+2/3 so winning 10-0 at 10+5 is better than losing 6-4 at 1+0 but worse than losing 5.5-4.5 at 1+0 that gives 91.125

We can give prizes to the first 3 players that are proptional to the number that they get in the ranking.

What is your opinion?

Joel Benjamin (SuperEasy) consistently scores 9, 9.5, or 10 out of 10 at 8'3" (not on your list but about same as 7'4"), so he would get between 65 and 91 points each time based on past results. I think it would be an interesting contest, but are you talking about an open contest on the internet, where cheating would be a concern, or a monitored event by you in Israel? It is unfortunate that any time money prizes are offered to the general public for similar events, cheating seems to be a real problem. It's fairly easy to catch cheaters in bullet chess events, where the cheating is generally automated, but in Rapid it's extremely hard to catch sporadic cheaters. Assuming a monitored event, I think most players will select the shortest time control at which they expect to score like 90% or so, which for most people means the maximum 10'5" will be chosen. If you get GMs then perhaps something like 5'3" will be popular. Results for the bot are pretty amazing, especially in the bullet range. Of course no one can compete at 1'0", but even at 2'0" performance of the bot is astronomical; against players above 2600 LiChess bullet it won 238 games vs. only 4 losses (ten draws). I don't think that incentives matter for bullet games, they are rarely drawn so it hardly matters if players resign in drawable positions, and they surely don't move too fast. But the longer the time limit, the more important incentive and motivation become, so I agree that if we want to measure performance in Rapid, some incentives are needed.

I consider organizing this competition.

I plan to donate 3000 shekels to the prize money that is something near 800 dollars.
I wonder how many players can practically play at the same time(I may invite only limited number of players to the competition because I assume Leela cannot play too many games at the same time and I do not want Leela to perofrm worse because of playing too many games at the same time.

If it is possible it is better not to allow everybody to play against lc0 in the time cf the competition but only the players who play against it.

Good evening Mr. Uri Blass. If at some point they wanted me to be part of the humans in the competition, I would be willing to compete. My horse Catecan and I are. citizens of the world, and I would like to represent a common flag: "That of being human" riding through the roads of the
life. For my part always grateful.

Today, Sunday, April 13, 2025, the upward trend in human elo against LeelaQueenOdds has been crossed. There are no longer any humans with an elo lower than 2200 among the top 100 humans on the leaderboard. If we take number 100 "POTASKUNCHIK" as an example, we see that his bullet Lichess elo is 2243, his Blitz elo 2249, and his rapid elo 2247. The number of games played against the Robot is 149 under a time control of 5 minutes plus 3 seconds per move. I wonder: In a comparison with the Fide elo, what is the relationship between the elo on the table and the Fide elo? It can be seen in the table that the vast majority of humans appearing on the table are strong players, to the point that just a few years ago it would have been considered that they would never lose against a robot without the robot's queen. But that's not the case. The robot looks like a real beast.

[lkaufman]

I think that we need some competition with significant money prizes to see what humans can really do against LeelaQueenOdds without Qd1.

My idea is that every player choose the time control that he prefers that has to be of the time X minutes per game + Y seconds per game when X and Y are integers and Y is the closest to X/2 that means the following time controls are allowed
1+0.1+1,2+1,3+1,3+2,4+2,5+2,5+3,6+3,7+3,7+4,8+4,9+4,9+5,10+5

Every player play a match of 10 games against LeelaQueenOdds when the ranking is based on the cube of number of points divided by X+Y

For example losing 6-4 at 1+0 gives 4^3 points that is 64 and winning 10-0 at 10+5 gives 10^3/15 points that is 66+2/3 so winning 10-0 at 10+5 is better than losing 6-4 at 1+0 but worse than losing 5.5-4.5 at 1+0 that gives 91.125

We can give prizes to the first 3 players that are proptional to the number that they get in the ranking.

What is your opinion?

Joel Benjamin (SuperEasy) consistently scores 9, 9.5, or 10 out of 10 at 8'3" (not on your list but about same as 7'4"), so he would get between 65 and 91 points each time based on past results. I think it would be an interesting contest, but are you talking about an open contest on the internet, where cheating would be a concern, or a monitored event by you in Israel? It is unfortunate that any time money prizes are offered to the general public for similar events, cheating seems to be a real problem. It's fairly easy to catch cheaters in bullet chess events, where the cheating is generally automated, but in Rapid it's extremely hard to catch sporadic cheaters. Assuming a monitored event, I think most players will select the shortest time control at which they expect to score like 90% or so, which for most people means the maximum 10'5" will be chosen. If you get GMs then perhaps something like 5'3" will be popular. Results for the bot are pretty amazing, especially in the bullet range. Of course no one can compete at 1'0", but even at 2'0" performance of the bot is astronomical; against players above 2600 LiChess bullet it won 238 games vs. only 4 losses (ten draws). I don't think that incentives matter for bullet games, they are rarely drawn so it hardly matters if players resign in drawable positions, and they surely don't move too fast. But the longer the time limit, the more important incentive and motivation become, so I agree that if we want to measure performance in Rapid, some incentives are needed.

I consider organizing this competition.

I plan to donate 3000 shekels to the prize money that is something near 800 dollars.
I wonder how many players can practically play at the same time(I may invite only limited number of players to the competition because I assume Leela cannot play too many games at the same time and I do not want Leela to perofrm worse because of playing too many games at the same time.

If it is possible it is better not to allow everybody to play against lc0 in the time cf the competition but only the players who play against it.

Very nice, should be a very interesting event. Yes, it is possible to restrict play to a specific list of players for a specific period of time; I would need approval since the queen odds bot no longer runs on my personal computer, but I think they will approve that request. The bot is currently limited to 18 games at once. Performance only suffers if there are multiple fast games (bullet or fast blitz); if everyone is playing Rapid or slow blitz then 18 games should be okay since the computer moves quickly and will therefore typically only be thinking about 2 or 3 positions at a time. So unless you expect a couple GMs to play I don't think you need limit it, probably you don't have 18 candidates anyway. I think the bot is currently limited to 9 games per color but that is easy to change to up to 18 with Leela White only. Time of day might be an issue for the restriction as the host has to be available, I think he is in my time zone but not sure. Or perhaps it can be controlled from Europe, not sure about that. Depending on how soon the event is, there might be an upgrade before it, depending on how well it tests when ready. We have found a problem with endgame play that we might be able to fix with an upgrade.

[lkaufman]

Today, Sunday, April 13, 2025, the upward trend in human elo against LeelaQueenOdds has been crossed. There are no longer any humans with an elo lower than 2200 among the top 100 humans on the leaderboard. If we take number 100 "POTASKUNCHIK" as an example, we see that his bullet Lichess elo is 2243, his Blitz elo 2249, and his rapid elo 2247. The number of games played against the Robot is 149 under a time control of 5 minutes plus 3 seconds per move. I wonder: In a comparison with the Fide elo, what is the relationship between the elo on the table and the Fide elo? It can be seen in the table that the vast majority of humans appearing on the table are strong players, to the point that just a few years ago it would have been considered that they would never lose against a robot without the robot's queen. But that's not the case. The robot looks like a real beast.

I believe your above statement is wrong. Player number 10 on the list, "Catecan", the most active player of all, has 2092 as his highest LiChess rating. I believe you are familiar with "Catecan", he is a human with an elo below 2200 among the top 100! Well, perhaps you meant to say "other humans"? As for comparing with FIDE elo, the main problem is that many players on the list are not choosing a strategy designed to get them a high place on the list. You and I and Joel Benjamin all choose what we believe to be roughly our best strategies, but many players may be unaware of the list, or have no idea of what time control works best, or simply have other reasons for playing the bot. I suspect that many players play it to improve their bullet skills, even though they lose almost every game doing so. So some GMs have low ratings on the list due to insisting on 1'0" bullet play, but would probably be quite high if they played blitz instead and always gave Leela White, assuming they played a hundred games or more. I may be number 7 on the list, but I am far from being the seventh best blitz player on the list. If you are correct that all the players (other than you) are above 2200 LiChess blitz (maybe some are only over 2200 in bullet?), then it would seem that on average the list is close to Lichess Blitz, which is perhaps 200 to 300 elo inflated relative to FIDE (200 based on stats of titled players, but they are a preselected group of players good at classical chess, so the real gap is likely larger). So perhaps subtracting 300 from the list is the best estimate for FIDE rating on average, but emphasis on "on average". Yes, even one year ago the idea of a bot beating (over 50%) "par" FMs in 3'2" blitz giving queen odds seemed preposterous, yet here we are. I should mention that only one knowledgeable person in my memory ever even mentioned the idea that some day a bot might even be able to give queen odds to masters; that was Bob Hyatt, over a decade ago! Everyone else (including me) thought this was ridiculous. Bob, if you read this, congratulations, it seems you were right, at least with respect to blitz chess!

[Uri Blass]

I think that we need some competition with significant money prizes to see what humans can really do against LeelaQueenOdds without Qd1.

My idea is that every player choose the time control that he prefers that has to be of the time X minutes per game + Y seconds per game when X and Y are integers and Y is the closest to X/2 that means the following time controls are allowed
1+0.1+1,2+1,3+1,3+2,4+2,5+2,5+3,6+3,7+3,7+4,8+4,9+4,9+5,10+5

Every player play a match of 10 games against LeelaQueenOdds when the ranking is based on the cube of number of points divided by X+Y

For example losing 6-4 at 1+0 gives 4^3 points that is 64 and winning 10-0 at 10+5 gives 10^3/15 points that is 66+2/3 so winning 10-0 at 10+5 is better than losing 6-4 at 1+0 but worse than losing 5.5-4.5 at 1+0 that gives 91.125

We can give prizes to the first 3 players that are proptional to the number that they get in the ranking.

What is your opinion?

Joel Benjamin (SuperEasy) consistently scores 9, 9.5, or 10 out of 10 at 8'3" (not on your list but about same as 7'4"), so he would get between 65 and 91 points each time based on past results. I think it would be an interesting contest, but are you talking about an open contest on the internet, where cheating would be a concern, or a monitored event by you in Israel? It is unfortunate that any time money prizes are offered to the general public for similar events, cheating seems to be a real problem. It's fairly easy to catch cheaters in bullet chess events, where the cheating is generally automated, but in Rapid it's extremely hard to catch sporadic cheaters. Assuming a monitored event, I think most players will select the shortest time control at which they expect to score like 90% or so, which for most people means the maximum 10'5" will be chosen. If you get GMs then perhaps something like 5'3" will be popular. Results for the bot are pretty amazing, especially in the bullet range. Of course no one can compete at 1'0", but even at 2'0" performance of the bot is astronomical; against players above 2600 LiChess bullet it won 238 games vs. only 4 losses (ten draws). I don't think that incentives matter for bullet games, they are rarely drawn so it hardly matters if players resign in drawable positions, and they surely don't move too fast. But the longer the time limit, the more important incentive and motivation become, so I agree that if we want to measure performance in Rapid, some incentives are needed.

I consider organizing this competition.

I plan to donate 3000 shekels to the prize money that is something near 800 dollars.
I wonder how many players can practically play at the same time(I may invite only limited number of players to the competition because I assume Leela cannot play too many games at the same time and I do not want Leela to perofrm worse because of playing too many games at the same time.

If it is possible it is better not to allow everybody to play against lc0 in the time cf the competition but only the players who play against it.

Very nice, should be a very interesting event. Yes, it is possible to restrict play to a specific list of players for a specific period of time; I would need approval since the queen odds bot no longer runs on my personal computer, but I think they will approve that request. The bot is currently limited to 18 games at once. Performance only suffers if there are multiple fast games (bullet or fast blitz); if everyone is playing Rapid or slow blitz then 18 games should be okay since the computer moves quickly and will therefore typically only be thinking about 2 or 3 positions at a time. So unless you expect a couple GMs to play I don't think you need limit it, probably you don't have 18 candidates anyway. I think the bot is currently limited to 9 games per color but that is easy to change to up to 18 with Leela White only. Time of day might be an issue for the restriction as the host has to be available, I think he is in my time zone but not sure. Or perhaps it can be controlled from Europe, not sure about that. Depending on how soon the event is, there might be an upgrade before it, depending on how well it tests when ready. We have found a problem with endgame play that we might be able to fix with an upgrade.

I think to tell humans 30 days before the event about the event in order to allow preperation time for humans who want to prepare so I guess the event is not going to be before the second half of May.

I do not know if more than 18 players are going to register with prize money when there is no limitation for rating of the players and I also consider to have a special prize of 500 shekels for a surprising player who does not win one of the first 3 players.and you can win this prize even by 5-5 at 10+5 based on difference in rating between expected rating based on your place in the competiion and your rating.

Example:
Suppose you get place 5 in the competition that is not enough for a prize but your ranking based on rating is worse.

I look at the rating of the 5th best player in the competition and reduce your rating from it and it is the difference I mean.

[Father]

Today, Sunday, April 13, 2025, the upward trend in human elo against LeelaQueenOdds has been crossed. There are no longer any humans with an elo lower than 2200 among the top 100 humans on the leaderboard. If we take number 100 "POTASKUNCHIK" as an example, we see that his bullet Lichess elo is 2243, his Blitz elo 2249, and his rapid elo 2247. The number of games played against the Robot is 149 under a time control of 5 minutes plus 3 seconds per move. I wonder: In a comparison with the Fide elo, what is the relationship between the elo on the table and the Fide elo? It can be seen in the table that the vast majority of humans appearing on the table are strong players, to the point that just a few years ago it would have been considered that they would never lose against a robot without the robot's queen. But that's not the case. The robot looks like a real beast.

I believe your above statement is wrong. Player number 10 on the list, "Catecan", the most active player of all, has 2092 as his highest LiChess rating. I believe you are familiar with "Catecan", he is a human with an elo below 2200 among the top 100! Well, perhaps you meant to say "other humans"? As for comparing with FIDE elo, the main problem is that many players on the list are not choosing a strategy designed to get them a high place on the list. You and I and Joel Benjamin all choose what we believe to be roughly our best strategies, but many players may be unaware of the list, or have no idea of what time control works best, or simply have other reasons for playing the bot. I suspect that many players play it to improve their bullet skills, even though they lose almost every game doing so. So some GMs have low ratings on the list due to insisting on 1'0" bullet play, but would probably be quite high if they played blitz instead and always gave Leela White, assuming they played a hundred games or more. I may be number 7 on the list, but I am far from being the seventh best blitz player on the list. If you are correct that all the players (other than you) are above 2200 LiChess blitz (maybe some are only over 2200 in bullet?), then it would seem that on average the list is close to Lichess Blitz, which is perhaps 200 to 300 elo inflated relative to FIDE (200 based on stats of titled players, but they are a preselected group of players good at classical chess, so the real gap is likely larger). So perhaps subtracting 300 from the list is the best estimate for FIDE rating on average, but emphasis on "on average". Yes, even one year ago the idea of a bot beating (over 50%) "par" FMs in 3'2" blitz giving queen odds seemed preposterous, yet here we are. I should mention that only one knowledgeable person in my memory ever even mentioned the idea that some day a bot might even be able to give queen odds to masters; that was Bob Hyatt, over a decade ago! Everyone else (including me) thought this was ridiculous. Bob, if you read this, congratulations, it seems you were right, at least with respect to blitz chess!

Thank you, Mr. Larry Kaufman.
Yes, it's true, the "anti-computer" strategy is essential for climbing the ladder. Perhaps the maxim that says, "The exception confirms the rule" applies here. And the exceptions could be so great that perhaps those who design anti-computer strategies or plans could eventually be candidates for advanced positions and even fight for leadership. So, it seems that a thorough understanding of the psychology of machines is a requirement for success. In this sense, Professor Bob Hyatt "prophesied" the future of chess-playing machine power; the professor, with a "mathematical" foundation, was able to "see" from "logic" the advent of the "probable future." Today I wonder: What will the probable future of "LeelaQueenOdds" be? What will be "the necessary future"? As for the "free future" of AI, I absolutely lack the faculties and intelligence to even imagine it. From the moment "free will" exists in machines, even "in the absence of consciousness," human beings will face a very serious survival challenge. Today I was happy to learn that, at least for now, I hold first place in the number of games played against the current robot.

[lkaufman]

I think that we need some competition with significant money prizes to see what humans can really do against LeelaQueenOdds without Qd1.

My idea is that every player choose the time control that he prefers that has to be of the time X minutes per game + Y seconds per game when X and Y are integers and Y is the closest to X/2 that means the following time controls are allowed
1+0.1+1,2+1,3+1,3+2,4+2,5+2,5+3,6+3,7+3,7+4,8+4,9+4,9+5,10+5

Every player play a match of 10 games against LeelaQueenOdds when the ranking is based on the cube of number of points divided by X+Y

For example losing 6-4 at 1+0 gives 4^3 points that is 64 and winning 10-0 at 10+5 gives 10^3/15 points that is 66+2/3 so winning 10-0 at 10+5 is better than losing 6-4 at 1+0 but worse than losing 5.5-4.5 at 1+0 that gives 91.125

We can give prizes to the first 3 players that are proptional to the number that they get in the ranking.

What is your opinion?

Joel Benjamin (SuperEasy) consistently scores 9, 9.5, or 10 out of 10 at 8'3" (not on your list but about same as 7'4"), so he would get between 65 and 91 points each time based on past results. I think it would be an interesting contest, but are you talking about an open contest on the internet, where cheating would be a concern, or a monitored event by you in Israel? It is unfortunate that any time money prizes are offered to the general public for similar events, cheating seems to be a real problem. It's fairly easy to catch cheaters in bullet chess events, where the cheating is generally automated, but in Rapid it's extremely hard to catch sporadic cheaters. Assuming a monitored event, I think most players will select the shortest time control at which they expect to score like 90% or so, which for most people means the maximum 10'5" will be chosen. If you get GMs then perhaps something like 5'3" will be popular. Results for the bot are pretty amazing, especially in the bullet range. Of course no one can compete at 1'0", but even at 2'0" performance of the bot is astronomical; against players above 2600 LiChess bullet it won 238 games vs. only 4 losses (ten draws). I don't think that incentives matter for bullet games, they are rarely drawn so it hardly matters if players resign in drawable positions, and they surely don't move too fast. But the longer the time limit, the more important incentive and motivation become, so I agree that if we want to measure performance in Rapid, some incentives are needed.

I consider organizing this competition.

I plan to donate 3000 shekels to the prize money that is something near 800 dollars.
I wonder how many players can practically play at the same time(I may invite only limited number of players to the competition because I assume Leela cannot play too many games at the same time and I do not want Leela to perofrm worse because of playing too many games at the same time.

If it is possible it is better not to allow everybody to play against lc0 in the time cf the competition but only the players who play against it.

Very nice, should be a very interesting event. Yes, it is possible to restrict play to a specific list of players for a specific period of time; I would need approval since the queen odds bot no longer runs on my personal computer, but I think they will approve that request. The bot is currently limited to 18 games at once. Performance only suffers if there are multiple fast games (bullet or fast blitz); if everyone is playing Rapid or slow blitz then 18 games should be okay since the computer moves quickly and will therefore typically only be thinking about 2 or 3 positions at a time. So unless you expect a couple GMs to play I don't think you need limit it, probably you don't have 18 candidates anyway. I think the bot is currently limited to 9 games per color but that is easy to change to up to 18 with Leela White only. Time of day might be an issue for the restriction as the host has to be available, I think he is in my time zone but not sure. Or perhaps it can be controlled from Europe, not sure about that. Depending on how soon the event is, there might be an upgrade before it, depending on how well it tests when ready. We have found a problem with endgame play that we might be able to fix with an upgrade.

I think to tell humans 30 days before the event about the event in order to allow preperation time for humans who want to prepare so I guess the event is not going to be before the second half of May.

I do not know if more than 18 players are going to register with prize money when there is no limitation for rating of the players and I also consider to have a special prize of 500 shekels for a surprising player who does not win one of the first 3 players.and you can win this prize even by 5-5 at 10+5 based on difference in rating between expected rating based on your place in the competiion and your rating.

Example:
Suppose you get place 5 in the competition that is not enough for a prize but your ranking based on rating is worse.

I look at the rating of the 5th best player in the competition and reduce your rating from it and it is the difference I mean.

I would expect that there will be a meaningful upgrade to the Queen odds bot before late May, there could also be a hardware upgrade if the 5090 processor becomes generally available. If so that might mean a 50% increase in the number of simultaneous games we can support.

[Father]

" I would expect that there will be a meaningful upgrade to the Queen odds bot before late May, there could also be a hardware upgrade if the 5090 processor becomes generally available. If so that might mean a 50% increase in the number of simultaneous games we can support. "

Mister Larry Kaufman: Joel's consolidation at the top of the leaderboard looks very strong, and he's making steady progress past the 3000 rating mark... Today I wonder: Will LeelaQueenOdds be able to reach world champion status in a blitz tournament against humans? Is that asking too much?

[lkaufman]

" I would expect that there will be a meaningful upgrade to the Queen odds bot before late May, there could also be a hardware upgrade if the 5090 processor becomes generally available. If so that might mean a 50% increase in the number of simultaneous games we can support. "

Mister Larry Kaufman: Joel's consolidation at the top of the leaderboard looks very strong, and he's making steady progress past the 3000 rating mark... Today I wonder: Will LeelaQueenOdds be able to reach world champion status in a blitz tournament against humans? Is that asking too much?

Right now, LeelaQueenOdds playing White is probably a tossup in 1'0" bullet against the top 3 bullet players (Nakamura, Carlsen, Firouzja). Perhaps with the next upgrade and a hardware upgrade to 5090 it will be too strong for any human at 1'0" bullet, but still an underdog at 1'1" or 2'0". If by "blitz" you mean 3'0", then I could imagine that it might challenge these top three in a couple years or so; I find it hard to imagine it ever winning a match against the top 3 at 3'2" where it can't expect to win games on time. But who knows, I never thought it would reach the present level! It has played over 100 games at 3'2" against GM Mikulas Manik (the one who lost his first five games to it) and the bot is still slightly ahead, but he is probably at best around "par" IM level in blitz now, a long way from the top guys.

[M ANSARI]

Is there a way to download that version of Leela to play on your own hardware? It really is a PIA to try and get games and from reading the latest posts now, it seems that if you do get to play QueenOdds BOT, it is not playing at full power?? Would be nice to have the BOT to play on your own hardware (like a 4090 Ti).

I did notice a few times the BOT plays one move blunders ... sometimes right out of the opening ... and is quickly lost (rare but happens). I am guessing that is because it is sharing hardware resources. I definitely feel it is not playing consistently strong ... sometimes it is crazy strong ... sometimes it feels not so strong.