Computer regressed?

( I was white )

Yes I checked it out now I can easily beat stockfish 4 and 5. Even draw with Stockfish 6. It becomes easier

I did a bit of spelunking through the code and the closest change I could find is this:
https://github.com/niklasf/fishnet/pull/115

Which is an attempt to solve:
https://github.com/ornicar/lila/issues/6482

This was made back in June, though, so wouldn't explain any kind of change in AI play over the last few days. If anyone is super motivated and has some free time they could always set up the developer environment and tweak variables in fishnet.py, etc

My guess is the change was made months ago and only being noticed now. Maybe I'm wrong, but I can't find any evidence in the code that anything else has changed related to Stockfish recently.

Another coincidence is that Stockfish 12 was released in Sept... but lichess uses a fork and not the official release (from what it looks like -- someone can confirm this?)
https://github.com/niklasf/Stockfish/tree/fishnet

(Even if Stockfish 12 were being used, the release notes claim it's even tougher.)

This is not how a 2000-rated player plays. Not even 1200!

SF5 gives up two pawns in an opening and then a free knight on like move 14. I beat it down in 50 moves and I was taking 2-5 seconds per move.

Doesn't chess.com have levels of stockfish? I'm just wondering: what do they do so that the range of rating is from 800 to GM without a problem?

lichess has stockfish 11 for ana lysis and max 8 for playing

Yeah, @A_0123456 i saw ur game and it didn’t take your free knight. Just that’s like level 1 player
( beginner )

Why do we think that level 5 is 2000 Elo? Here's the code from Fishnet:
https://github.com/ddugovic/fishnet/blob/9ae101aca092c3aaa8ab109fb24ab3e9c84f4e6f/fishnet.py

LVL_ELO = [800, 1000, 1250, 1500, 1750, 2000, 2300]
LVL_MOVETIMES = [50, 100, 150, 200, 300, 400, 500, 1000]

That's 0 indexed, so level 5 is 1750. This only has 7 entries b/c it looks like level 8 is a special case in the code (altho I'm not familiar with it so I could always be wrong -- just trying to find some signal in the noise).

There were changes... back in June. You can see them here:
https://github.com/ddugovic/fishnet/commit/731cd89f6b57f8ee0fce221399d6b79d708335b7#diff-87894c890485bfc244f1cd5ebe6a9de7

The major assumption here is that Fishnet controls AI games and not just analysis (which it looks like, but maybe someone can confirm). If anyone else wants to look deeper look at the bestmove(self, job) method along with the Elo params to understand how it's interacting with core Stockfish. HTH.

In my game Stockfish 5 made A LOT of blunders too. See below.