Contexto Unlimited
Find the secret word. Meaning matters; spelling doesn’t.
How to play Contexto
- Enter any English word to see its semantic rank.
- A smaller rank means your guess is used in contexts more like the secret word.
- Follow the warmest ideas until you reach rank #1.
This is not a spelling game. If the answer were forest, words such as tree, woodland, or jungle might guide you closer even though their letters are different. This page deals another puzzle whenever you want one; the shared once-a-day challenge with its streak lives at Contexto Daily.
How the ranking works
Each puzzle uses one fixed semantic word map built from GloVe vectors. We precompute the complete ranking for each puzzle and send it as a compact static data file. That keeps every player on the same ranking while guesses respond instantly in the browser.
A rank measures contextual association, not a dictionary family tree. A broad word may be very close because it frequently appears beside the answer, while one particular example can occur in quite different writing. Treat a warm guess as a direction to explore, not proof that every narrower word must rank higher.
Frequently asked questions
What does a rank mean?
Rank #1 is the answer. Rank #20 is closer than rank #2,000; the absolute similarity score is intentionally hidden so the ordering stays easy to read.
Do hints count as guesses?
No. A hint makes a large jump while you are far away, then takes progressively smaller steps inside the top 1,000 so the final stretch still requires deduction. Hints are labeled separately in your list.
Why did my plural or past tense become a base word?
Answers use standard dictionary forms. Clear inflections such as animals, told, and eaten are shown and ranked as animal, tell, and eat. Ambiguous words such as saw and left keep their own meaning and rank.
What are database1, database2, and database3?
The small code shown after your first guess identifies the fixed semantic map for that puzzle; it never changes midway through a game. database1 uses Stanford’s 2014 GloVe Wikipedia + Gigaword model, database2 uses the updated 2024 WikiGiga 300-dimensional model, and database3 combines both precomputed rankings with Reciprocal Rank Fusion. For the technical background, read the original GloVe paper, A New Pair of GloVes, and the RRF paper.