Bonus challenges

Solved all 100? Nice work. Here are some directions to take it further if you finish early. None of these are required, so pick whatever looks fun.

Feel free to use AI for anything down here. The main challenge asked you to go solo, but the bonus round is wide open, so reach for an LLM or coding assistant as much as you like.

Optimize for a target average

Set yourself a target average score and tune your solver until you hit it. Pick a number you think is reachable, then iterate on your opening word and strategy to get there. Lower averages are harder, so see how far down you can push it.

Just keep your solve rate at 100% while you do it. A great average doesn't count for much if you're failing puzzles to get there, so make sure you're still solving all 100 games.

Solve it in hard mode

Can you still solve all 100 with hard mode turned on? In hard mode every hint a guess reveals has to be reused by your later guesses: a green letter stays locked to its position, and a yellow letter must appear again. Add ?hard=1 to the request:

curl -X POST 'https://wordle.rdme.io/game/1?hard=1' \
  -H 'Content-Type: application/json' \
  -d '["SOARE","PIZZA"]'

If a guess breaks one of those rules the request fails with 400 and a message saying which constraint was violated.

Bring in an LLM

Can an LLM beat your deterministic solver? There are two angles here, and you can try either or both:

Start every game with a random word

What if each game had to open with a random word instead of your favorite opener? Call GET /random before each game and use whatever it returns as your first guess. It hands back one random word from the ~12,970 valid Wordle guesses:

curl 'https://wordle.rdme.io/random'
{ "word": "CRANE" }

Add ?format=text to get just the bare word.

Find the optimal first word

Using the full answer list from GET /words, work out which opening word gives you the best head start. Crunch the letter and position frequencies, measure how many answers each candidate rules out on average, and rank openers by the average number of guesses they'd need across all 2,315 puzzles. The endpoint returns every possible answer as a JSON array:

curl 'https://wordle.rdme.io/words'
["ABACK", "ABASE", "ABATE", "ABBEY", "ABBOT", ...]

Add ?format=text to get them newline-separated instead.

Build a Wordlebot

Go a step further and grade a whole game, like the New York Times' own WordleBot. For each guess, count how many answers were still possible before and after (the "outs" you eliminated), then turn that into two scores:

Report a per-guess breakdown and an overall skill/luck rating for the game, and you've essentially rebuilt WordleBot. The mockup on the right is the kind of recap you're aiming for, so click through it to see the pieces.

Or do your own thing

None of these grabbing you? Go off-script. If there's some other angle you'd rather chase, like a wild opener, a visualizer, or a totally different strategy, we're down for whatever you want to build. Surprise us.


Back to the docs