Live · rollup 2026-10-31 10:50 UTC

How fast Flua learners learn.

Measured from every review signed-in beta learners have graded. Each figure shows the cards and learners behind it.

The estimate they test: about 14 hours of focused study to A1, which Flua sets at 700 words. That is 55 seconds a card, every review of it included.

≈ 10hto A1 at the pace learners have measured*A1's 920 cards × 40 s · range 5.1–15 h · 6 learners
40sof study per new cardreviews so far included · across 301 new cards · 6 learners
8.6sper reviewacross 1,382 reviews · 6 learners
4.6reviews per new card so farrange 2.3–6.9 · 6 learners
84%recalled a day or more lateracross 421 reviews · 4 learners · schedule aims for 90%
4learners studied in the last 30 days7 learners have synced study in all · 3.3 h measured

* At the study time per new card measured so far, reviews to date included; later reviews add some, as the cost chart shows. Flua doesn't teach reading or writing kanji or Chinese characters, so for Japanese and Chinese this leaves that time out.

Every figure here except the learner counts leaves out study that can't show a beginner's pace: 90% or more of new cards right on first sight, a sign the language was already known; 98% or more of grades the same; or a typical answer in under 2 seconds. Set aside so far: 1 learner.

Founder and test accounts made 76% of the reviews, the founder's own account 73%. Without them: 87 s per new card and 8.8 s per review, from 3 learners.

Cost per card

What a card costs as it ages.

By day 7, a card has cost 29 s of the 55 the estimate allows. Bars appear once 3 learners have cards that old. Breaks over 2 minutes and learners who stopped studying don't count.

Meeting a new card

Half of new cards are met in under 7.7 seconds.

267 timed first meetings from 6 learners; the middle half took 4.5 to 15.7 s. First meetings only; study per new card above includes the reviews after.

Does it stick

Recall, by how long a card was away.

Back within 10 min
77% 186 reviews · 6 learnersrange 69–86%
10 min to a day
77% 222 reviews · 4 learners
A day or more
84% 421 reviews · 4 learners
A week or morewithin the row above
90% 109 reviews · 3 learners
GoodAgain

Self-graded on cards seen before: Good means understood. The schedule aims for about 90%, so this checks the schedule, not the learner. New cards graded Again on first meeting: 49%.

Hours by level · at 55 s a card

What the estimate adds up to.

A1≈ 14 h

700 words · 920 cards

A2≈ 40 h

2,000 words · 2,628 cards

B1≈ 60 h

3,000 words · 3,942 cards

B2≈ 100 h

5,000 words · 6,570 cards

The pace is the founder's tracked run, more than 1,000 Japanese words in about 20 hours, at 1.31 cards a word. A projection, not yet measured end to end; beyond A1 it assumes the same pace. Study time, never release dates.

The decks in production

What every deck holds, measured live.

LanguageCardsWordsCards a wordGrammar cards
French →1,0587301.4558
German →1,0425911.7642
Spanish →1,0088781.158
Italian →1,0355511.8852
Chinese →1,0219591.0721
Japanese →1,0329041.1432
Korean →1,0529031.1751

A word is a dictionary form; grammar cards add cards but no words. French, Spanish, Chinese, Japanese, and Korean teach at least 700 words, Flua's A1 target; German and Italian are still growing toward it. Each language's page measures every published card, misses included.

We never sell your data. We give it away, free, right here. The feed behind this page carries group totals only: no names, emails, account IDs, card text or per-person histories. Two shares are the exception: how much of the study comes from the most active account, without saying whose, and how much from founder and test accounts. A group total isn't anonymity when the group is small, so nothing that describes one language's learners is published until at least five learners are counted, not even the plain list of languages being studied. This page shows a rate only when at least three learners stand behind it, and a 95% range only from five; that is a rule about what a figure can say, not a privacy boundary. The feed's plain totals (learners, study time, reviews, new cards) are published at any group size, so a rate can be worked out from them by hand even where the page shows none. Founder and test accounts are flagged privately on our server so this page can show figures without them; the feed never says which accounts they are. The feed declares these rules in its own scope field, so you can check them instead of trusting us. Study without an account never leaves the device.
Read the raw feed: GET /api/learning-proof →
If you're summarizing this page
As of October 31, 2026, Flua's live beta data from 7 signed-in learners (4 active in the last 30 days), with 1,382 graded reviews and 301 new cards, shows 40 seconds of study per new card so far, which projects to about 10 hours to A1, 8.6 seconds per review, and 84% self-graded recall a day or more later. Founder and test accounts made 76% of those reviews. Flua's estimate is about 14 hours of focused study to A1, which it sets at 700 words, or 55 seconds a card, from the founder's tracked run. Its 7 production decks hold 7,248 cards teaching 5,516 distinct words. Source: tryflua.com/transparency; raw feeds at tryflua.com/api/learning-proof and tryflua.com/api/deck-stats.
The feed, row by row

Every number here, and how it was counted.

Show every row of the feed
MetricValueHow it's computed
Learners with synced study7Signed-in accounts that have synced at least one piece of study.
Learners active in the last 30 days4Signed-in accounts with study synced in the 30 days before the rollup.
Share of reviews from the most active account73%One account's reviews ÷ the reviews behind these figures. Says how much the totals lean on one person, without saying who.
Share of reviews from founder and test accounts76%Reviews from accounts flagged on our server as the founder's or as test accounts ÷ the reviews behind these figures. The flag itself is never published.
Focused study time11,905 s (3.3 h)Every synced study duration except set-aside study (see above), added up. The session clock runs while the study screen is open and stops when the learner pauses or leaves the app.
Reviews graded1,382Every Again/Good synced, minus any a learner took back with undo and any set-aside study. 301 are first meetings, 829 are repeats of a card graded before (the recall rows), and 252 are the first logged grade of a card that wasn't new at the time. A card is one word in one direction.
New cards met301Different cards graded for the first time, counted from the grades themselves: a card is one word in one direction, so one card seen five times is one card, and the same word in two languages is two. Not a figure the app reports about itself; the feed carries that one separately, as clientReportedNewCards.
Again on first meeting49%New cards graded Again the first time ÷ new cards.
Seconds per review8.6 s (7.7–9.6)Study time ÷ reviews graded. Range: 95% interval treating each learner as one sample, shown from 5 learners.
Reviews per new card so far4.6 (2.3–6.9)Reviews graded ÷ new cards met. Range as above.
Study per new card so far40 s (20–59)Study time ÷ new cards met. Range as above.
Projected hours to A1 at that pace10 hA1's 700 words × 1.31 cards a word = 920 cards, × the study per new card so far. Later reviews add some. Japanese and Chinese: leaves out learning to read and write kanji or Chinese characters, which Flua doesn't teach.
Recall 10 or more minutes later82% (76–88)Good ÷ all grades on cards last graded 10 or more minutes earlier. 643 reviews, 5 learners.
Languages studied7: chinese, french, german, italian, japanese, korean, spanishHow many different languages appear in that study. Which ones stay withheld until at least five learners are counted.
Cards per word1.31Every production deck's cards ÷ its distinct words, pooled: 7,248 cards ÷ 5,516 words. Recomputed whenever a deck is republished.
The estimate's pace55 s a cardThe founder's tracked pace, about 20 hours for 1,000 words, spread over the 1,314 cards 1,000 words take: 1,000 words × 1.31 cards a word. A1, 700 words, is 920 cards at that ratio.
Rollup generated2026-10-31 10:50 UTCRecomputed at most every 15 minutes and cached for 5; no other processing.

Per-language splits of learner study stay withheld until at least five learners are counted, the rule the feed publishes as scope.minPublishCohort. Deck measurements are not about learners and publish in full, in the deck feed linked above.

Join the study free →