Mate in 1
Benchmark page
Grok 4.20 Beta scored 53.3% strict accuracy on ChessBench, solving 80 of 150 fixed Lichess tactics when the answer had to match the expected UCI move line exactly.
Strict accuracy
53.3%
Correct puzzles
80 / 150
Prompt tokens
91,129
Total cost
$7.90
Benchmark updated
Mar 25, 2026, 10:01 PM
Find the tactical line that creates a fork.
Find the line that wins an undefended piece.
Solve a forced mate in one move.
Solve a forced mate in two moves.
Find the tactical line that exploits a pin.
This page gives Google a stable, canonical URL for the model and describes the benchmark result in plain HTML instead of only inside the interactive explorer.
It also links back to the benchmark methodology and exposes model-specific copy, numbers, and sample puzzle outcomes that are relevant to searches like "Grok 4.20 Betachess benchmark" and "Grok 4.20 Betatactical reasoning results".
For the full dataset definition and scoring details, visit the ChessBench dataset and methodology page.
These examples add readable, model-specific context beyond a single leaderboard number.
Mate in 1
Mate in 1
Mate in 1
Mate in 1
Mate in 1
Expected line: b1h7
Model line: No parseable line returned
Parse status: invalid_count
Mate in 1