Hello, recently YaneuraOu level 2, 3 and 4 have been replaced by stockfish, but level 5 is YaneuraOu again
As YaneuraOu old level 8 was better than, I believe, any human, as there are new levels 9 and 10, does this mean that the level 2, 3 and 4 of fairy stockfish aren't replacing Yaneura level 3 and 4 (level 2 being broken) but they are just there to have more bot variety ? Is YaneuraOu new level 5 equivalent to YaneuraOu old level 3 ?
I've played many times against the old YaneuraOu level 3 and the level 5 right now is crushing me, it seems to be way better than the old 3 ... is it just me ?
Regarding the AI changes on Lishogi, Wanderer had already clearly explained this in the update announcement back at the end of 2025(lishogi.org/blog/post/aUhtgBEAAC0ABLJI) —specific details can be found under the "Level Difficulty" section, that part was written out plainly and frankly, many questions could be resolved just by giving it a little thought.
Right after Lishogi’s AI updated, I tried a few games myself and easily defeat YaneuraOuLV5 at the time. But soon after, its strength noticeably increased, which tells me the team has been dynamically adjusting these leveled AIs behind the scenes and make their strength fluctuates. What you observed with the old YaneuraOuLV2 being broken was likely also due to such adjustments not landing quite right.
It’s worth mentioning other strictly level-capped AIs (like Piyo engine LV1–20), in order to show their “restricted” nature, they often deliberately make blunders visible even to the naked eye—low-level piece drops for no reason、meaningless repetition、or reckless aggressive moves. When training against them, there’s a fat chance they simply hand you an advantage, which diminishes the training value. In contrast, the new YaneuraOu LV5 basically avoids those kinds of mistakes even a beginner would spot; its style is solid and orthodox. Personally I think it’s very well suited for practice——only by genuinely defeating an AI that “seems” flawless on the surface can you feel your actual skill progressing.
Another AI change on Lishogi shows up at the top end:For research purposes I occasionally use a local GUI engine stronger than Lishogi’s analysis board to challenge YaneuraOu LV10 (screw platform rules). Doing so, I noticed that even my locally running strong engine sometimes shows slightly passive against LV10 at the beginning. It makes me strongly suspect YaneuraOuLV10 comes with a strong, refined opening book. Of course that’s just speculation, plus, even if it does have one, the positions the AI handles well may not suit human players to copy directly, so whether to learn or imitate LV10’s openings is really a matter of personal judgment. Also, post-game or when submitting kifu, the platform offers a quick one-off computer analysis—this got faster after the update, taking only about 30 seconds to draw the evaluation curve and flag blunders and mistakes, which definitely makes review more convenient. However, that speed obviously trades off precision. I’ve submitted some high-level game records where YaneuraOu temporarily flagged a move as a mistake; but when I opened the analysis board and gave it more time to think deeply, the eval quickly stabilized with minimal fluctuation, proving the move was actually fine—just a brief misjudgment from rushed analysis. So it still holds useful reference value for average players, but for Godly Players (around 7-dan and above), independent thinking should remain primary, combining their own understanding of shogi principles with critical judgment.
The reason I’m writing such a long reply is simply because I’m passionate about studying AI and topics like “New Bot Level” naturally pull me into detailed discussion. Yet when I bring this up in communities around me, people repeatedly advise focusing my energy purely on real PVP matches and the issues exposed there. They see little point in bot training, even encouraging beginners fresh off learning the rules to jump straight into online play. When I personally engage in PVP, I do find myself tensing up almost involuntarily, factors like countdown pressure magnify negative effects, pushing me toward instinctive moves over deep calculation, leading to sloppy plays and frequent blunders. Maybe it’s precisely in that kind of high-pressure, messy, even ugly struggle that you find the practical training no AI can ever replace…
You can't post in the forums yet. Play some games!