Skip to main content

🧱 Cementix 09/04/2026

tries: 0
Semantic temperature
🧊
🌋
FrozenColdWarmHotBurning

Find the mystery word. Score from 0 to 1000 ‰ — the closer you get, the hotter the bar.
↑/↓ history · Enter to submit

🌡️ My tries
Start with a general word…
nature, work, animal, love…
# Word Score

Cémentix: the semantic word game powered by artificial intelligence

Cémentix is the most intellectually distinctive game in the Brekiz collection. Unlike spelling-based games (Wordus, Hangman) where you either know the word or you do not, Cémentix operates on meaning. Every guess is evaluated by a word2vec model — a neural network trained on millions of French texts — which measures the semantic distance between your proposed word and the hidden target. The closer to 100 your score, the more closely related your word is in meaning to the answer.

How word2vec works: the technology behind Cémentix

Word2vec is a neural network architecture developed at Google in 2013 that transforms words into numerical vectors in a high-dimensional space. The key insight: words that appear in similar contexts in large text corpora end up with similar vector representations. "King" and "queen" have similar vectors. "Dog" and "puppy" have similar vectors. "Summer" and "vacation" have similar vectors — because they co-appear frequently in texts about the same topics.

Cémentix uses a word2vec model trained specifically on French texts — news articles, books, Wikipedia, web content. The similarity score you see (0 to 100) is a rescaled cosine similarity between your proposed word's vector and the target word's vector. Cosine similarity measures the angle between two vectors in that high-dimensional space: angle of 0° = identical direction = score of 100, angle of 90° = perpendicular = score near 0.

This means Cémentix is not testing your vocabulary, but your mental model of how words relate to each other in meaning — a fundamentally different cognitive skill from any other word game.

Interpreting your Cémentix scores

Understanding what each score range means is crucial for developing good strategy:

  • Score 0–20: completely unrelated. Your word is in a completely different semantic domain. Abandon this direction entirely.
  • Score 20–40: very distant but not random. There is a faint thematic connection, but you are likely in a neighboring topic. Try to identify what vague connection your word might have to the answer.
  • Score 40–60: same general domain. You are in the right neighborhood. The answer is likely in the same topic category (sports, cooking, music, nature, etc.).
  • Score 60–80: closely related. You are getting warm. The answer and your word likely co-occur in many similar contexts. Look for more specific terms within the same concept cluster.
  • Score 80–99: very close. The answer is likely a synonym, a strongly associated concept, or a word in the same very specific domain. You might be guessing near-synonyms or the same word family.
  • Score 100: you found it. The word is the target.

The funnel strategy: how expert players approach Cémentix

Step 1 — Domain mapping (attempts 1–5)

Your first 5 guesses should cover completely different semantic domains: try one word from each major category — nature (tree, water, mountain), human activity (work, music, sport), abstract concepts (time, love, power), objects (car, house, book), and living things (animal, person, plant). The scores across these probes will rapidly triangulate which semantic neighborhood the target lives in.

Step 2 — Domain exploration (attempts 6–20)

Once you have identified the right domain (say, everything related to cooking scored above 40), explore that domain systematically. Try specific words: chef, recipe, ingredient, oven, pan, flavor, sauce, bake, cuisine... Each score refines the exact corner of the semantic space the target occupies.

Step 3 — Precision hunting (attempts 21+)

When scores reach 70+, you are in the hot zone. Now is the time to try synonyms, related compound concepts, and adjacent vocabulary. If "kitchen" scores 75, try "cooking," "culinary," "gastronomy," "recipe," "dish," "preparation." The target will emerge as the word that makes semantic sense at the center of the high-scoring cluster.

Cémentix vs other semantic word games

GameLanguageModelHints providedAvg. attempts
Cémentix (Brekiz)Frenchword2vec (French corpus)Score 0–100 per guess25–50
CemantixFrenchword2vec (French corpus)Score + rank proximity30–60
SemantleEnglishword2vec (English corpus)Score + "getting warmer"40–100+
ContextoMultipleContext-based modelPosition in ranked list20–40

Cémentix's key distinguishing feature within Brekiz: it is one of 10 games, not the only daily game. Your Cémentix score contributes to your global daily ranking and league points, creating stakes that make each session count beyond pure curiosity.

Learning French vocabulary through Cémentix

One of the less-discussed benefits of Cémentix: it builds genuine language intuition. Because the game forces you to think about semantic relationships rather than spelling rules, regular play develops a more nuanced mental lexicon. Players who play Cémentix daily for several months consistently report improved French vocabulary in contexts outside the game — reading, writing, and conversation — because they have been forced to map semantic networks deeply and repeatedly.

This makes Cémentix particularly valuable for French learners at intermediate and advanced levels. Unlike vocabulary flashcard apps that reinforce isolated word-definition pairs, Cémentix trains contextual semantic relationships — which is how native speakers actually organize vocabulary in memory.

Frequently asked questions about Cémentix

Why does a synonym sometimes score lower than an unrelated word?
This happens because word2vec measures distributional similarity (how often words appear in similar contexts), not definitional similarity. Two words can be synonyms in dictionary terms but appear in very different contexts in actual text corpora. For example, "commence" and "begin" are synonyms, but "begin" appears in more casual texts while "commence" appears in more formal/administrative contexts — so their vectors may diverge more than expected.

Are proper nouns (names, cities) accepted in Cémentix?
It depends on the word being in the model's vocabulary. Common proper nouns that appear frequently in French texts (Paris, Einstein, Shakespeare) are in the model and produce meaningful similarity scores. Rare proper nouns may not be recognized and will return an error or zero score.

Does the Cémentix model understand conjugated verbs and plurals?
The model handles common inflected forms (plurals, conjugated verbs) as separate vocabulary entries. However, the target word is always in its base form (infinitive for verbs, singular for nouns). Guessing a conjugated form of the correct verb may score high but not reach 100. Trying the infinitive is recommended when you have identified the likely verb.

What is the record for the fewest attempts to find the Cémentix word?
Some players have found the target in as few as 1–3 attempts on very lucky days where their first broad probe happened to hit the exact semantic neighborhood. The "Cémentix Instinct" badge is awarded for finding the word in under 5 attempts — achieved by less than 5% of players on any given day.

Is Cémentix easier or harder than Cemantix?
Difficulty is subjective and varies by day depending on the target word. Both games use similar technology but different word corpora and word selection algorithms. Many players find them similarly challenging on average, with specific days being harder on one or the other. Playing both is a great way to cross-train your semantic intuition.