Sudoku 101

Prompts can have a huge impact on LLM performance, both in research and in real-world applications.

That’s why Renbu is hosting its first-ever prompt engineering competition, challenging participants to solve Sudoku puzzles using nothing but prompt engineering.

Everyone competes on a level playing field. All participants use the same model, Claude Haiku 5.5, with the same effort and temperature settings.

To keep the competition fair, submissions are limited to 3 per day. Make the most of them! Download the dev set to test your prompts locally, or try them in the Playground, before submitting.

Ready to put your prompt engineering skills to the test? Submit your prompt and see if you can claim the top spot on the leaderboard!

The model gets your prompt as its only message, with {puzzle} replaced by the puzzle's grid. If your prompt has no {puzzle}, the grid goes first, then a blank line, then your prompt. Nothing else is added: not the rules, no instructions and no answer format, so your prompt has to explain the task and ask for an answer we can read.

The grid and nothing else: one row per line, cells separated by spaces, empty cells as dots, and "|" and "-" lines marking the boxes. Symbols are 1–9, then A–G on grids bigger than 9×9. Grids can be 4×4, 6×6, 9×9, 12×12 or 16×16; the Data tab shows the mix. The rules aren't included, so say what you need the model to know.

For example:

. . | 1 2
1 . | 4 3
----+----
2 . | 3 .
4 3 | . 1

Put the completed grid inside <answer></answer> tags: one row per line from top to bottom, the symbols of a row separated by single spaces. Only a complete grid that keeps every given and follows the rules counts as solved.

For the puzzle above:

<answer>
3 4 1 2
1 2 4 3
2 1 3 4
4 3 2 1
</answer>

• We read the last <answer> block. Without one, we take the last complete grid anywhere in the reply: it's scored the same, but counted as a format deviation.
• A row can also be written without spaces (1234), with a label such as "Row 3:", or with | , ; [ ] ( ) between symbols. Separator lines such as ----+---- are skipped.
• Letters can be either case, and on grids bigger than 9×9 the numbers 10–16 are read as A–G.
• . 0 _ ? x count as blank cells, and a grid with a blank isn't solved.
• Text inside <think>, <thinking>, <reasoning> or <analysis> tags is ignored.
• A reply cut off by the output token limit, or still running at the timeout, scores as not solved.