Your evaluation as black being much better seems to be correct. However, I would argue that neither of the evaluations you provided is accurate.
As for the chess.com evaluation, I think your result is incorrect due to a network/server error of some sort because I can't seem to reproduce your results (Edit: Discrepancy in result is due to game being drawn at this position as OP has clarified in comments). Here's my result of chess.com's engine at depth = 20 using Stockfish 16.1 Lite: (-3.43)
For Chessvision, using the Chrome extension, I got -6.2 for this position using Stockfish HCE at depth 20 (which is similar to your reported evaluation of -5.9)
To compare these results, let us consider few more evaluations: (click on the image to view in full size)
- Lichess stockfish 11 using HCE gave the evaluation of
-6.8 at depth = 20.
- Lichess stockfish 16 using NNUE gave the evaluation of
-3.5 at depth = 20.
- And finally, Lichess stockfish 16 using NNUE gave the evaluation of
-4.0 at depth = 30
In order to understand the reason for difference in these evaluations, one must know the difference between Stockfish HCE, Stockfish NNUE and Stockfish Lite.
Stockfish HCE was basically using handcrafted evaluation functions to evaluate a given position and did not actively learn from games played in similar positions. See this recent question to understand the pitfalls of handcrafted evaluation.
Then came Stockfish NNUE (Efficiently Updatable Neural Network in reverse) which use neural nets to add learning capability to the above handcrafted evaluation function. Introduction of NNUE has made HCE model obsolete (Since July 2023, HCE model has been removed by Stockfish)
Finally, chess.com's Stockfish Lite is in the process of transitioning from HCE model to smaller NNUEs. (link)
Hence, the most accurate engine evaluations among all of the above at this point in time would be Stockfish 16 using NNUE at depths 20 (-3.5) and 30 (-4.0).
