A June 1, 2026 publication describes a compact, 800,000-parameter loop transformer that it says achieved 100% on Sudoku-Extreme after 15 minutes of training.
The publication, dated June 1, 2026, presents a compact loop transformer with 800,000 parameters for structured puzzles. It says the model reasons in a way similar to a SAT solver and reached 100% on Sudoku-Extreme after 15 minutes of training.
That result is what the publication reports; the available summary does not specify the evaluation protocol, dataset, or comparison conditions. To verify the claim, consult the original material and check how Sudoku-Extreme, training, and the 100% metric were defined. If you use AI to study or apply the approach, do not submit confidential organizational data without first applying your organization’s data-protection policies.