14.d5! was a great Tal sac, and Stockfish 12 finds that it's even better than Tal thought. His annotations say 14...♗xf1 15.dxe6 ♘c5! 16.exf7+ ♔f8 16.exf7+ ♔f8 17.♘e5 ♘xb3 18.♕xb3 ♕d6 19.♘c5! (Tal says "with a very strong attack"; an understatement, because the computer says =9!). An even prettier is the Tal-like line 17.♘xc5 ♕xd1 18.♘e5! winning; not bothering to recapture the ♕ but Δ mate on either d7 or g6.

Computer disagrees with Tal's statement 15.♖e1 dxe4 16.♖xe4 ♘f6 "with chances for both sides". It continues 17.♕xd8+ ♖xd8 18.♖e3 with a clear advantage for white (~+2) because Black can't complete development. E.g. 18... ♔f8 19.♘e5 ♘d5 20.♖f3 ♗f6 21.♘xc6 ♖d6 22.♘xa7 ♗e2 23.♖f5, with an extra ♙ and Black's ♔♖ still out of the game.

Computer says Black's best was 16...♗xf1, but one frightfully complex line is 17.♘xc6 ♕c7 18.♕d4 ♗b5 19.♖e1 ♗xc6 20.d6 ♕b8 21.dxe7 ♘f6 22.♘c5 h5 with about equal chances although White is a ♖ down.

Tal sys that 18... ♗d6 looks like a blunder but is White's best move. The computer finds 18...♘c5 as equally good, but this requires a whole exchange sac 19.♘c6 ♖xc6 20.♗xc6 ♘d3 21.♖e2 ♗g5 22.♖c2 with equal chances because of White's uncoordinated pieces esp. ♘a4 vs Black's centralized ones.

In Informator, Tal considered 19...♗xh2+ 20.♔xh2 ♕h4+ 21.♔g1 fxe6 22.♗xe6+ ♔h8 23.f3! and claimed a clear advantage. Stockfish 12 finds that Black can force a draw by perpetual check with 23...♗b7 24.♗xd7 ♗xf3! 25.♕d2 (25.gxf3 ♕g3+ 26.♔h1 ♖c4 27.♗g4 ♕h4+ 28.♔g1 ♖xg4+ 29.fxg4 ♕g3+) 25...♗xg2 26.♔xg2 ♖c4 27.♗h3 ♖f3 28.♖e8+ ♔h7 29.♔xf3 ♕xh3+.