Semantle Junior: How to Play, Score & Win

Semantle Junior is the small, friendly version of the semantic guessing game. Same idea as its parent: you type words, and the game tells you how close each one is in meaning to a secret word you cannot see. What changes is the deck. Junior's answers come from a much smaller, much more everyday vocabulary, which makes it the right place to learn the mechanic before you attempt the main board.

This page explains the machinery rather than the answer. Today's Semantle Junior word is not printed anywhere below. What you get instead is a method: how to read the score, how to ladder from a broad category down to a specific object, why a hot guess can still be wrong, and how to practise so that the skill transfers.

What Semantle Junior actually is

Semantle Junior is a daily word game built on word embeddings. Every English word in the game's vocabulary has been converted into a vector — a long list of numbers that encodes the kinds of sentences the word tends to appear in. Two words that show up in similar company end up with similar vectors, and the game measures the angle between them to produce a similarity score.

You guess a word. The game scores it. You guess again. There is no limit on guesses, which is the first thing that separates this family from Wordle-likes: you cannot lose by running out of rows, because there are no rows. You can only lose by giving up or by spending an evening circling the wrong neighbourhood.

The furniture around the game is modest. There is a hint button, a give-up button, a proximity column that appears once you are close, an archive of past puzzles so you can catch up or practise, and optional stats that can sync across devices if you log in. The game was created by David Turner and is now part of The Word Finder, which also hosts the archive.

One practical detail trips up new players every day: the puzzle does not roll over at your local midnight. It runs on UTC, which lands at 8pm in New York the evening before. If you sit down at 11pm expecting a fresh word and get yesterday's, you are not early — you are on the wrong clock.

How the similarity score works

The score runs from 100 down to roughly minus 34 in practice. A score of 100 means your word is the secret word, and that ends the game. The theoretical floor is minus 100, but real English words almost never go below the low thirties negative, because even unrelated words share some context in a news corpus.

The single most important thing to understand is what the number measures. It does not measure synonymy. It measures shared context — whether two words turn up in similar sentences. The classic formulation is John Rupert Firth's: you shall know a word by the company it keeps. That definition has consequences that feel wrong until you internalise them.

Antonyms score high. Hot and cold appear in the same sentences constantly, so their vectors sit close together. If you guess "hot" and the score jumps, the answer might be "cold". This is not the game being broken; it is the game being honest about how language works.

Polysemy distorts everything. A word with two meanings gets one vector that blends them, weighted by which meaning is commoner in the corpus. "Patent" leans toward the legal sense, not the leather-finish sense, so "leather" scores further away than a human would expect. When a guess scores bizarrely low, ask which sense of the word the corpus probably favours.

Part of speech matters unevenly. A noun guess against a verb answer can score poorly even when the concepts are related. If a whole run of nouns is scoring flat, try the verb form of your best noun, or the noun form of your best verb.

Capitalisation is not cosmetic. Typing a capitalised word can change how the game treats it, and some mobile keyboards capitalise automatically. If a guess returns an unexpected score or no ranking at all, check that the word went in lowercase before you conclude anything about meaning.

The rank counts down, and that changes how you read the board

Once your guess lands within the thousand closest words, the game stops showing only a decimal and starts showing your position in that neighbourhood. Junior counts down: the nearest word to the answer sits at 1, and the edge of the neighbourhood sits near 1000. The game's own notes describe this as the biggest format change from the main board, made because new players find a number that decreases toward the target more intuitive than one they have to interpret.

Two habits follow. First, treat the rank as a distance, not a score: 400 to 120 is progress, and 120 to 118 is not worth celebrating. Second, notice when the rank stops moving while the score creeps up. That pattern means you are refining within a neighbourhood instead of moving between neighbourhoods, and it is the signal to change category rather than to keep polishing.

If you have never played the main game, the practical difference is small. If you play both, do not carry rank intuitions across: the two boards have different vocabularies, so the same rank number means a different distance.

Why Junior's answers are more concrete

The Junior word list has an origin story that explains almost everything about how it plays. It was assembled by a five-year-old boy and an eight-year-old girl shouting out words, then shuffled. The main game's list was built a different way, and the game itself notes that any overlap between the two is coincidental.

Children name things. They name animals, foods, household objects, vehicles, body parts, toys, weather. They do not name "inflation", "jurisdiction" or "ambiguity". So Junior's answers cluster where a child's vocabulary clusters: concrete nouns and very common terms. The game states this directly — Junior's secret words are typically nouns or very common terms.

That has three strategic consequences.

Abstract openers are wasted. Opening with "idea", "system" or "concept" tells you almost nothing, because the answer is unlikely to live in that region. Opening with "house", "water" or "dog" puts you in the middle of the answer distribution.

Category ladders are short. In the main game you may need six rungs to descend from "entity" to a specific object. In Junior, three rungs often land you within a hundred places.

Everyday knowledge beats sophisticated knowledge. The player who can list forty things in a kitchen will outperform the player who has read more widely but thinks in abstractions. This is a game about the physical world.

Semantle vs Semantle Junior

The compression point deserves a sentence of its own. Because Junior's pool is smaller and its words are commoner, scores bunch up. A 45 on the Junior board is not the same achievement as a 45 on the main board; the same number means less. Read your score against your own best guess of the session, and read the rank against the pool you are actually playing.

Starting words that suit Junior

Your opening guess is not a guess at the answer. It is a calibration: it tells you which region of the vocabulary the answer lives in. For that purpose, pick a word that is central to everyday physical life, not a word that is merely common in text.

Three anchors is a better opening than one, because a single score cannot tell you direction. Guess one word from the home, one from the natural world, and one from food or animals. Compare the three scores, and you have a rough map.

Once one anchor clearly outscores the others, stop calibrating and start descending. The mistake is to keep firing anchors after the signal is clear — three more anchors will not tell you anything the third one did not, and each one costs a minute.

The category ladder method

The ladder is the core skill of this game, and it is the same skill in both versions. You move from a broad category to a narrow one, one rung at a time, and you never skip a rung. Each rung should score better than the last; when it does not, you have taken a wrong branch and must climb back to your last good word.

Two things make the ladder work. The first is that rungs 1 and 2 are cheap and reliable: they are almost always scored high enough to confirm you are in the right domain. The second is that rung 4 is where most games are won, because Junior's answers are instances. Hypernyms feel productive and score well without ever being the answer, which is exactly the trap described in the next section.

There is a discipline to the ladder that separates fast players from slow ones: one variable at a time. If you jump from "kitchen" to "kettle" and the score drops, you cannot tell whether you left the right setting or whether the class was wrong. Move from "kitchen" to "appliance" first. If that rises, you now know the setting was right and the class was right, and "kettle" becomes a much better bet.

When a rung fails, do not guess another instance immediately. Climb back to the last rung that scored well and descend a sibling branch: from "appliance" to "utensil", from "utensil" to "container". The vocabulary is a tree, and a wrong instance tells you only that you chose the wrong leaf, not that the branch is dead.

A final refinement for Junior specifically: the ladder is often only three rungs deep. Because answers are concrete, "domain → setting → instance" frequently lands within a hundred places. If you find yourself on rung five in Junior, you have probably skipped past the answer and are now describing it rather than naming it.

Why "hot but wrong" happens

A large score from a word that cannot be the answer is the most confusing thing in the game, and it is not random. Four mechanisms explain nearly every case.

Hypernymy. "Furniture" scores well against "chair" without being a chair. Broad terms sit near their instances in the vector space, so they read as hot. The fix is to descend, not to celebrate.

Antonymy and co-occurrence. "Cold" and "hot" share sentences. So do "buy" and "sell", "doctor" and "patient", "win" and "lose". A hot guess can be the opposite pole of the answer rather than a neighbour of it. When a run feels contradictory, test the opposite deliberately.

Polysemy. The answer's vector may reflect a sense you are not thinking of. "Spring" is a season, a coil and a water source, and its vector splits across all three. If the neighbourhood makes no sense, list the other senses of your best word and try one of those.

The neighbour is not the target. Words that appear in identical collocations are near-duplicates in the space — "puppy" and "dog", "sofa" and "couch", "plate" and "dish". Being one step from the answer feels like a 90 because the vector distance is tiny, but you still have to name the actual word.

The practical response to all four is the same: when a guess is hot, do not guess more words in that neighbourhood at random. Ask which relationship you are looking at — instance, opposite, other sense, or synonym — and then make one guess that tests that relationship specifically.

How to read the score number

There is no official table of score zones, and treating one as gospel will mislead you. What follows is a rough reading that helps you decide whether to stay or move, not a set of thresholds.

Two rules keep this honest. Compare every guess to your personal best, never to 100: going from 8 to 31 is a large move even though 31 looks small. And watch the derivative rather than the level: three guesses that each add two points mean you are polishing, while one guess that adds fifteen means you have changed neighbourhood, which is the only movement that matters.

Negative scores are not a failure state. They are information: you have found a region of the vocabulary that is actively far from the answer, which is often more useful than a vague 12, because it rules out a whole branch.

Practice that actually transfers

Unlimited guesses make this game easy to play and hard to improve at, because nothing forces you to be efficient. Deliberate practice fixes that.

Play the archive with a guess budget. Pick a past puzzle and give yourself fifteen guesses. The constraint forces the ladder instead of the scattergun. Review where your fifteenth guess landed compared with your fiftieth from a normal run.

Run single-variable sessions. Spend a session guessing only hypernyms, then a session guessing only instances. You will learn which one your instincts overuse, and most players discover they are hypernym-heavy.

Log your guesses and scores. A log makes plateaus visible. Players routinely believe they are climbing when the last six guesses moved four points total.

Do reverse runs. After a puzzle ends, start a fresh run with a word you know sits near the answer and try to reach it in five guesses. This teaches the shape of the space faster than any amount of blind exploration.

Review your give-ups. When you surrender, spend two minutes asking which rung you skipped. The answer is nearly always the same: you jumped from a setting to an attribute without testing the class in between.

If you want a second daily rep in the same family, the main board is the natural next step, and our Semantle hints and strategy page covers the differences in depth. Players who want a different flavour of the same skill can also try Contexto, which ranks guesses by similarity too but without the score display.

Spoiler-free hints for today's puzzle

Hints in this game are unusual, because the useful hint is almost never a letter. It is a position in the vocabulary. A good hint narrows the region without naming the word; a bad hint is either so vague it removes nothing or so specific it ends the game.

The ladder below is the one to use on yourself, or to ask a friend for. Take one rung at a time and stop as soon as you can name three candidates.

1. Class of thing. Is it alive, an object, a place, a food, or an action? This is the cheapest hint and it removes the most.
2. Setting. Where would you meet it — a kitchen, a garden, a street, a classroom, a body?
3. Size and function. Is it bigger than a hand? Do you use it, eat it, visit it, or watch it?
4. Shape of the word. How many letters, and does it start with a common consonant cluster? This is the last rung, because it converts a vocabulary hunt into a word hunt.

Notice what is missing: the first letter as an opening hint. In a game scored by meaning, the first letter is nearly worthless until you know the region, and it tempts players into scanning a dictionary instead of thinking about the world.

The game also has its own hint button. Use it late rather than early. A hint spent at guess two is a hint spent before you know which question to ask, and the button does not refund.

Common mistakes

1. Guessing synonyms of a hot word. Synonyms cluster, so a run of them produces small score changes and no progress. Move up a rung and across.
2. Staying on a hypernym. "Furniture" will never be the answer in Junior. If your best score is a category word, you are one rung too high.
3. Ignoring part of speech. A flat run of nouns against a verb answer looks like a wrong domain. Test the verb form before abandoning the region.
4. Treating the score as a percentage. It is a cosine similarity. 50 does not mean "half solved", and 80 does not mean "one letter to go".
5. Skipping the calibration phase. One anchor cannot give you direction. Three anchors from different domains cost three guesses and save thirty.
6. Playing on the wrong clock. If the puzzle looks identical to yesterday's, you are early, not stuck. The rollover is UTC, not local midnight.

Frequently asked questions

What is Semantle Junior?

Semantle Junior is the easier version of Semantle, a daily word game where you guess a hidden word and are scored on how similar each guess is in meaning rather than in spelling. Its answers come from a much smaller, more everyday vocabulary, typically concrete nouns and very common terms, and it gives you unlimited guesses.

How is Semantle Junior different from Semantle?

Three differences matter. Junior uses a much smaller and more concrete word list, so answers are everyday objects and common terms rather than abstract vocabulary. Junior counts the proximity rank down so the nearest word sits at 1. And because its pool is smaller, scores bunch together, so the same number means less on Junior than on the main board.

What does a similarity score of 100 mean?

A score of 100 means your guess is the secret word, which wins the game. Scores run down to roughly minus 34 in practice. The number measures how similarly two words are used in sentences, so opposites such as hot and cold often score close to each other.

Why did a wrong word score so highly?

Usually one of four reasons. The word may be a broader category that contains the answer, an opposite that shares sentences with it, a different sense of a polysemous word, or a near-synonym that appears in identical contexts. Test which relationship you are looking at before guessing again.

What time does Semantle Junior reset?

The new word arrives at midnight UTC, which is 8pm in New York the evening before. That is why the puzzle can feel a day ahead or a day behind if you expect it to change at your local midnight. The archive lets you play any earlier puzzle.

Does this page show today's Semantle Junior answer?

No. This page is a strategy and hints guide and does not publish the current word. It teaches the category ladder, score reading, starting words and a four-rung spoiler-free hint ladder that you can apply to any day's puzzle.

Related: Semantle hints and strategy · Play Semantle free · Contexto hints today · All daily word games