VARC
Reading, against the clock
Passage-grouped questions with a 60-second reading window before the timer starts charging.
Timed at90s per question
Ninjatest is the AI-native arena for the world's toughest entrance exams — a real-time, 1v1 ranked mode where every question is matched to your level, scored by the second, and rated so you always know exactly where you stand.
Built for GMATCATGRELSATJEEUPSCNEETMCATSATUCATACTTSAGMAT prep battles
One arena for the world's toughest competitive exams
Every battle is a compressed mock: three VARC, three DILR, three Quant — or nine from the section you fear most.
VARC
Passage-grouped questions with a 60-second reading window before the timer starts charging.
DILR
Puzzle sets served as a group — the hardest section gets the longest clock.
QUANT
The highest speed multiplier of the three. A fast correct answer out-scores a slow one.
Scoring
Every answer is scored server-side. Client clocks and client claims are ignored.
Speed bonus
Bonus accrues in five-second blocks. A random guess is worth exactly zero expected points.
Live
Real-time presence: you know the moment your opponent locks in — never what they chose.
Spectate
Any live match is watchable, read-only, with answers hidden until the reveal.
VARC
Passage-grouped questions with a 60-second reading window before the timer starts charging.
DILR
Puzzle sets served as a group — the hardest section gets the longest clock.
QUANT
The highest speed multiplier of the three. A fast correct answer out-scores a slow one.
Scoring
Every answer is scored server-side. Client clocks and client claims are ignored.
Speed bonus
Bonus accrues in five-second blocks. A random guess is worth exactly zero expected points.
Live
Real-time presence: you know the moment your opponent locks in — never what they chose.
Spectate
Any live match is watchable, read-only, with answers hidden until the reveal.
VARC
Passage-grouped questions with a 60-second reading window before the timer starts charging.
DILR
Puzzle sets served as a group — the hardest section gets the longest clock.
QUANT
The highest speed multiplier of the three. A fast correct answer out-scores a slow one.
Scoring
Every answer is scored server-side. Client clocks and client claims are ignored.
Speed bonus
Bonus accrues in five-second blocks. A random guess is worth exactly zero expected points.
Live
Real-time presence: you know the moment your opponent locks in — never what they chose.
Spectate
Any live match is watchable, read-only, with answers hidden until the reveal.
Ninja — the AI layer
Ninja reviews your match the way a good teacher would: where you lost time, which trap you took, what to drill next. Then it builds the practice set itself.
A per-question breakdown of your time against your opponent's — including the two questions that actually decided the match.
You won the match in DILR: 41 seconds faster across the set, both correct. Q4 is where it turned — your opponent spent 68s on the seating arrangement and guessed; you skipped in 12s and banked the clock.
Question-level reasoning whenever you want it — locked during live matches, open the second you finish.
Why is option B wrong here? The passage seems to support it.
B restates the author's example, not the argument. The passage uses the 2010 census as an illustration — the claim itself is in the final paragraph, which only D paraphrases without adding "always".
Weakness-targeted sets composed from your match history. If arrangements keep costing you, arrangements are what you get.
The bot plays at your rating, honestly — it answers on its own clock and never peeks at the answer key.
NINJA-BOT
plays at 1240 ELO · unrated
Bot is reading the question…
0:47Tap play. The queue pairs you with someone at your level — the band widens the longer you wait, so you always get a game.
Derived from the live queue rule: band = min(1000, 100 + 20 × seconds waited).
Time to match
under 10s
typical at peak
band widens 20 ELO per second
Rated fairness
zero-sum
every point won is a point lost
100 ELO floor, no farming
Rematch guard
3 per day
same rated pair
so ratings stay honest
Every dial — clocks, penalties, ELO — is tuned so the rating means what it says.
Everything a rated battle needs, checked by the server — never the client.
Answer fast, score more. Guessing is EV-neutral by construction — the penalty rides the same speed curve.
max per question, every section
Tuned per section, capped per question, measured server-side.
The whole loop takes minutes — the rating sticks.
01
One tap. The server finds your equal.
02
Nine synchronized questions. Same clock, same order, no pauses.
03
Margin-weighted, zero-sum ELO. Beat a stronger player, gain more.
Join the waitlist — early aspirants get founding badges and first access to rated seasons.
Welcome back.