stockfish does not think during your time. It has max 0.4 seconds for each move (at level 8) and that's it.
The thing is that on your PC, stockfish is only playing with you. On lichess, it handles everybody at the same time.
Might you consider easing the depth limits at the higher levels? From what I understand, Stockfish sometimes goes very deep into the evaluation between, say, two moves that are already early on obviously better than others.
By the way, I think the skill level code in Stockfish itself (separate from the depth and time limitations implemented here) could sometimes produce odd results. The meat is here, in the do_skill_level() function:
What happens is that if, for example, there is only one move that is any good, Stockfish would always pick that one, even if that move is to avoid or exploit some obscure tactic that happens to be available, even at the lowest skill level.
The depth limits for the AI on lichess seem to prevent that, though, when it comes to tactics, but from how I understand it, the above would still mean that Stockfish level 1 will simply never, for example, just drop a piece.
And from my experience as a bad player, it often doesn't matter how much better the one best move is -- I may well be able to miss the one decent move simply because it's only one move and there are so many others to choose from.
And the above would also mean that Stockfish level 1 will also simply never fail to snatch a dropped piece, so even against it you have to be pretty cautious through the whole game.
In my opinion level 8 plays 2000 +/- 400 elo. Sometimes really good like GM at its max, and sometimes it has its really bad moments.
At a kinda draw situation in this game http://lichess.org/analyse/sdam7skh it played 35...Nh4, which gave away the game. The rest was relatively easy for me to win.
Had like 3 unsuccessful tries before (2 loses and 1 draw), but on my fourth try I won. I would say I "can" play around 2000-2100, when my mind is clear (1800 +/- 300 when things occupy my brain), so I think it plays around 2000 +/- 400.
If you go too much into open fields, it will most-likely outplay you. But give it a pawn for some offensive, and you can claim a draw or even a win. With only 0.4 secs of time for thinking, it can barely judge if a pawn is worth taking or not.
As thibault mentioned the AI (level 5-8) have improved.
As such my personal opinion on the ratings has change too...
Level 1=1100 +-300
Level 2=1350 +-300
Level 3=1500 +-300
Level 4=1650 +-300
Level 5=1850 +-300
Level 6=2150 +-300
Level 7=2400 +-300
Level 8=2600 +-300
last two (lvl 7 and 8) are big guesses just based on ply count (dont have any tests on them done).
Hi, i have just beat the lvl 7 without using takeback of course... http://lichess.org/analyse/cxohcufm
Has Stockfish AI been altered?
I say that because I was struggling to beat level 6 before
And sorry for my english.
Legend, or thibault, if you have an answer...
you see Onche
something is bad happening with their Stockfish
in fact, the best chess engine in the world is Houdini (now version 3)
then Critter, then Stockfish, Deep Rybka,Deep Shredder,etc
my Houdini easily beats this Stockfish level 8, and Deep Shredder does the same, even in couple of moves that Stockfish played against Shredder, Stockfish made a few mistakes, (alerted by Critter and Houdini and Shredder)
So, this comp AI stockfish crap is actually...
A STUFF FULL OF CRAP CODE !
of course if thibault didn't turn off the permanent brain option...