FinalSudoku: 1,000 Extremely Hard Sudoku Puzzles for AI Benchmarking

FinalSudoku is a benchmark dataset for evaluating exact, verifiable puzzle-solving in AI systems. The public release contains 1000 extremely hard Sudoku puzzles, averaging an ER of 11.114 with verified answers. It is maintained by faranglab and was last updated on 2026-06-30.

Use Cases

• Benchmarking AI puzzle-solving performance based on the 1000 hard puzzles with verified answers.
• Training models for logical reasoning based on the 81-character string puzzle representation.
• Evaluating the generalizability of AI systems using the separate 3000-puzzle private evaluation set mentioned in the description.

Strengths

• Contains 1000 puzzles with verified answers, providing a concrete evaluation target.
• Puzzles are characterized as extremely hard, with an average ER of 11.114.
• Maintains a separate 3000-puzzle private set for official leaderboard results, suggesting a structured evaluation protocol.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the public set is specified, but overall dataset size and file formats are unknown.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

faranglab

Freshness

Last updated 2026-06-30 09:16:46; freshness should be verified.