A Simple Pangram Solver for NYT Spelling Bee

05/03/2026

nextjs openai spelling-bee tiny-win

A small Next.js app that finds valid NYT Spelling Bee pangrams from seven letters and a required center letter.

Hero image for A Simple Pangram Solver for NYT Spelling Bee

In early 2025, I made a small Next.js app for finding NYT Spelling Bee pangrams.

You enter the 7 letters, mark the required center letter, press one button, and (hopefully) get back a cleaned list of valid pangrams. The whole thing is intentionally small: one UI, one server route, one model call, and one local filtering step before results come back.

Mobile screenshot of the Pangram Solver app showing letters entered, center letter O selected, and JOCULAR returned as the pangram result. — The whole loop in one screen: letters in, one tap, pangram back.

The shape

Request flow diagram for the pangram solver: user input in the Next.js UI goes to a server route, which builds a prompt, calls gpt-5-mini, then filters and returns only valid pangrams. — Simple shape: form, server route, model call, then local filtering before anything comes back to the UI.

Model call, then local validation

The model handles the word-generation step: find real words that use only the provided letters, including the center letter, and use all 7 letters at least once. After that, the server rechecks every candidate with deterministic local rules before returning anything.

const instructions = "You are a word expert specialized in finding pangrams...";

const input = `Given letters "${letters}" with "${centerLetter}" as center letter...
- Each word must contain the center letter
- Use only the available letters
- Use all 7 letters at least once
- Return only words, one per line`;

const response = await openai.responses.create({
  model: "gpt-5-mini",
  instructions,
  input,
});

const content = response.output_text || "";
const words = content
  .split("\n")
  .filter((line: string) => line.trim().length > 0);

const pangrams = cleanAndFilterPangrams(words, letters, centerLetter);

The nice pattern here is descriptive prompt in, deterministic filter out.

That second step is what makes the app usable. The model can suggest candidates, but the final result still has to pass a straightforward local validation check.

UI states

There are really only two states that matter: before the solve and after the solve. I wanted both to stay simple. Very little chrome, one clear action, and a result list that is easy to scan.

Mobile screenshot of the Pangram Solver app in its empty state with blank inputs and the Find Pangrams button disabled. — Empty state: waiting for the 7 letters and the center letter.

Mobile screenshot of the Pangram Solver app showing a completed solve with JOCULAR listed under Pangrams Found. — Solved state: one valid pangram back after the server filters the model output.

That is basically it!

It was also one of those small early-2025 builds where I mostly described what I wanted, let an LLM produce the first pass, and then cleaned it up into something real in about an hour. Thanks for reading!