Skip to content

Flashcards

FSRS vs SM-2 vs Leitner: Which Flashcard Algorithm Wins?

FSRS vs SM-2 vs Leitner compared: how each flashcard scheduler decides when you review, what the evidence says, and which one fits how you study.

By iranimij Published 6 min read

FSRS is the most accurate of the three flashcard schedulers, SM-2 is the long-standing default behind Anki, and Leitner is the simplest. All three space your reviews so that you see a word just before you forget it. They differ in how much they know about each card, and that decides how many reviews you do for the same result.

This guide explains how each one decides “when next?”, what the evidence says, and which to pick for vocabulary.

What does a spaced repetition algorithm actually do?

It predicts when you are about to forget a card and puts the card in front of you at that moment. Spaced repetition means reviewing material at growing intervals instead of all at once. A review of 317 experiments found that the best gap between reviews grows with how long you need to remember the material (Cepeda et al., 2006), so a scheduler’s job is to grow the gaps sensibly. For why spacing beats cramming in the first place, see our piece on spaced repetition vs cramming.

How does the Leitner system schedule cards?

Leitner uses boxes with fixed review gaps. A correct answer moves a card to the next box, which you open less often. A miss sends it back to box 1. Sebastian Leitner described it in his 1972 book So lernt man lernen. We cover the full method in the Leitner system explained.

  • Strength: you can explain it in one minute and predict your workload.
  • Limit: every card in a box shares the same gap, whether the word is easy or stubborn.

How does SM-2 schedule cards?

SM-2 multiplies the previous gap by a number called the ease factor. Piotr Wozniak wrote the first SuperMemo program in 1987, and the SM-2 algorithm is described in his 1990 text (Wozniak, 1990). The first gap is 1 day, the second is 6 days, and after that each gap is the last one times the card’s ease factor, which starts at 2.5 and is not allowed to fall below 1.3. You grade each answer from 0 to 5, and the grade nudges the ease factor up or down. A grade below 3 restarts the card’s repetitions.

So an easy card might go 1, 6, 15, 37 days, while a hard card with an ease factor of 1.3 crawls: 1, 6, 8, 10. Anki’s default scheduler is based on SM-2 (Anki FAQ).

  • Strength: it adapts per card and has been tested by millions of learners.
  • Limit: it only looks at the last answer and a single ease number. It does not model forgetting, and it has no idea what retention you are aiming for.

How does FSRS schedule cards?

FSRS (Free Spaced Repetition Scheduler) models your memory of each card and schedules to a retention target you set. It tracks three values for every card: difficulty (how hard the card is for you), stability (how long the memory lasts) and retrievability (the chance you remember it right now). The ideas build on a research paper by Ye, Su and Cao, who framed scheduling as an optimization problem and tested it on a large dataset of real learner logs (Ye et al., 2022).

The setting you notice is desired retention: the share of cards you want to still remember at review time. In the FSRS4Anki tutorial, the allowed range is roughly 70% to 99% in recent Anki versions (FSRS4Anki tutorial). A value around 90% is a common starting point. Higher means more reviews and fewer lapses; lower means a lighter queue and more forgotten words.

As of Anki 23.10, Anki ships both algorithms, and its FAQ states that with FSRS you do fewer reviews to reach the same retention (Anki FAQ). FSRS can also be optimized on your own review history, which works better the more reviews you have logged.

Which one is more accurate: FSRS vs SM-2 vs Leitner?

FSRS has the strongest evidence, SM-2 sits in the middle, and Leitner is simplest but least adaptive. The most-cited comparison is the open Benchmark of Spaced Repetition Algorithms, built on thousands of real Anki collections. It reports that FSRS predicts whether you will recall a card better than SM-2 does for the large majority of collections.

Two honest caveats. First, the benchmark is run by the FSRS community, so it is not an independent trial. Second, SM-2 was never designed to output a probability of recall, so the benchmark has to add formulas to it before it can compare. Prediction accuracy is also not the same as “you learn more”. The benchmark measures how well a model forecasts recall, not language outcomes.

LeitnerSM-2FSRS
Year1972 (book)1987–19902022 (research paper)
Adapts to each cardNo (per box)Yes (ease factor)Yes (difficulty and stability)
You choose a retention targetNoNoYes
Setup effortVery lowLowMedium
Easy to predict your workloadYesSomewhatLess so

How does Radicard schedule words?

Radicard uses the simple end of this spectrum on purpose. It runs a seven-box Leitner system with review gaps of 1, 2, 4, 8, 16, 31 and 120 days, so you never touch a setting. The words come from photos: you snap a menu, a sign or a book page, and Gemini AI picks 5–15 words at your CEFR level, which are saved to your device and scheduled automatically. Reviews work offline. You can see how the boxes work in the Leitner section of the home page.

Its limits match Leitner’s: every word in a box shares one gap, so it does not tune itself to each card the way FSRS does. If you want that level of control and you are happy to manage a deck, Anki with FSRS is the stronger tool; our Anki vs Quizlet vs Radicard comparison goes through the trade-offs.

Which algorithm should you choose for vocabulary?

Choose the scheduler you will actually open every day. For most learners at A2–B2 level, the card quality and the daily habit matter more than the math. Use this as a rule of thumb:

  • Beginner, or you hate settings: Leitner, in an app that runs the boxes for you.
  • Anki user with a few hundred cards or more: switch FSRS on, keep desired retention near 90% and adjust only if the queue feels too heavy or too many words slip.
  • Already happy on SM-2: you do not have to move. A working routine beats a better algorithm you abandon.

Whichever you use, a weak card stays weak. Check our techniques for remembering vocabulary for how to write cards that stick.

What to do next

  1. Check which scheduler you are on: in Anki, open the deck options and look for the FSRS switch; in another app, look for “boxes” or “intervals”.
  2. If you use Anki, test FSRS on one deck for two weeks with desired retention at 90%, and compare your daily review count.
  3. If you use Leitner, keep the gaps growing (1, 2, 4, 8 days or more) and send missed cards back to box 1.
  4. Cap new cards at 10–20 a day so the review queue stays manageable.
  5. After a month, judge by two numbers only: how many reviews you do per day, and how many words you still recall.

Frequently asked questions

Is FSRS better than SM-2?

On the evidence available, yes for prediction and workload. Anki’s FAQ says FSRS needs fewer reviews than its default algorithm for the same retention, and the open benchmark finds FSRS predicts recall better than SM-2 for most collections. The benchmark is run by the FSRS community, though, so treat the size of the gap with some caution.

Is the Leitner system a spaced repetition algorithm?

Yes. It is the simplest kind: cards sit in boxes, correct answers move a card to a box reviewed less often, and a miss sends it back to box 1. The gaps are fixed per box, so it does not adapt to individual cards the way SM-2 or FSRS does.

What is desired retention in Anki's FSRS?

Desired retention is the share of cards you want to still remember when they come up for review. A common starting point is 90%. Raising it gives you more reviews per day, and lowering it gives you fewer reviews but more forgotten cards.

Should a beginner worry about which algorithm to use?

No. All three beat cramming, and consistent daily review matters far more than the scheduler. Choose the tool you will open every day, and revisit the algorithm only when your review pile or your forgetting rate starts to bother you.