Leela vs. Stockfish 12 — different chess engines, different analysis
I’ve been using chess engines since the early 1990’s. I started with one of the first (strong) software programs called MChess written for MS-DOS. The programs have come a long way, from being standalone hardware devices to modular engines that can be added to a graphical interface.
As of this writing, Stockfish 12 is considered the strongest chess engine in the world, rated 3558. There are others too. Komodo, Houdini, Leela and Fritz are all rated over 3000 ELO. The difference in playing strength is not apparent to the average human. Whenever I put a position for an engine to analyze, the recommendations are very similar. This changed recently when I gave the following position to Leela and Stockfish 12.
This is a position from the King’s Gambit Quade Variation. I am White and my opponent just played Qxe4. Leela thinks the position is relatively equal and that after Qd2, Black captures the knight on c3 and play continues.
Stockfish does not agree! It sees Kf2(!) as not only strong but winning. The idea being that White now threatens Re1 with a direct threat against the Black king. This makes sense to me. Qxe4 might actually be a dubious move since it opens the e-file which leads to trouble. See the analysis of both engines below.
After a lot of analysis, the conclusion is that Stockfish’s assessment of the position is correct. Kf2 is a multi-purpose move in that it gets the king out of danger and prepares a very strong counter-attack with Re1.
Getting a second opinion
I like to challenge moves suggested by chess engines. There are some lines that I often reject, not because they aren’t good but because they are not consistent with what the position calls for. For example, Stockfish sometimes recommends a random rook move that has nothing to do with what is happening for the battle in the center. After I give it move time to analyze, it comes back with a more realistic alternative.
For this position, Qd2 felt too passive for my tastes. Oddly enough, Leela considered it the only move for White to stay in the game. That’s when I decided to launch Stockfish. As soon as I did, the evaluation immediately swung positive and continued to grow. This is the only time in decades that I have ever witnessed this type of disparity between two chess engines.
I am curious what other people’s engine’s have to say. Can you put the above chess position on your favorite chess engine? What does the analysis say? Do you have examples of a similar disparity between two chess engines? Please share below!