← All articles
Paivio blog

Spaced Repetition Algorithm: How SM-2 and FSRS Work

Spaced Repetition Algorithm: How SM-2 and FSRS Work

Student reviewing a physical study card

A spaced repetition algorithm is the rule set that decides when a card comes back. It reads how well you recalled it, estimates how close you are to forgetting it, and schedules the next review just before that point. SuperMemo’s SM-2 does this with a simple multiplier. Newer models such as FSRS estimate your memory for each card from your review history. In both, the quality of the schedule depends on the quality of your ratings and your material.

This guide explains both in plain terms, shows what the research says about results for exam prep, and gives you practical settings for pre-med, medical and nursing coursework.

PaivioMake Spaced Reviews Fit Your StudyImport your Anki decks, notes, or topics into Paivio and build a personalized review system around your learning pace.Visit

TL;DR

  • A spaced repetition algorithm turns your recall ratings into the next review date for every card, so easy cards fade into the background and hard ones return sooner.
  • SM-2, the older SuperMemo-based method, grows intervals with a per-card multiplier. FSRS, the newer one, models memory strength and aims at a retention target you choose.
  • Studies of medical students link regular flashcard use to better board scores, though the evidence is mostly observational.
  • Honest ratings and well-made cards matter more than which algorithm you pick.
  • Paivio lets you import your own Anki decks and schedule reviews around your exam date.

Table of Contents

What a spaced repetition algorithm actually does

Every flashcard system has to answer one question: when should this card show up again? Without an algorithm you guess, and guessing leads to two failures. You review easy material too often and waste hours, or you let hard material slip past the point where recall is still possible.

The algorithm replaces the guess with a loop:

  1. You see a card and try to recall the answer.
  2. You rate how it went.
  3. The algorithm updates what it believes about that card.
  4. It sets the next review date.
  5. The card disappears until that date, then the loop repeats.

Three well-studied effects sit underneath. The forgetting curve says memory fades over time. The spacing effect says reviews spread over time stick better than the same effort packed together. The testing effect says pulling an answer out of memory strengthens it more than rereading. One academic review of the technique describes spaced repetition as harnessing all three, and reports that medical students preparing for their boards remembered nearly three times as much when they used it.

The optimal gap is not “as long as possible” or “as short as possible.” Work on optimizing spaced repetition schedules frames the ideal schedule in terms of recall probability, under a cost for reviewing too often. In practice, that means you want to review a card when you have a good chance of still remembering it, but have already lost some of the edge.

If you want the method behind the math first, our guide to the spaced repetition method covers how to use it day to day.

How recall ratings change the next interval

The rating is the only signal the algorithm gets from you. It cannot read your mind, so the button you press is the whole input.

Most systems use a short scale, such as Again, Hard, Good and Easy. The logic is consistent across them:

Your rating What the algorithm infers Typical effect on the card
Again (failed) You did not recall it The card goes back to a short interval and is treated as harder
Hard You recalled it with effort The interval grows only a little
Good You recalled it as expected The interval grows by the normal amount
Easy Recall was effortless The interval grows more than normal

A few consequences follow from this:

  • Inflated ratings damage the schedule. If you press Good on a card you guessed, the algorithm pushes it further out than your memory supports. You then meet the card at a point where you have forgotten it.
  • Rating is about retrieval, not recognition. Decide your answer before you flip the card. If you only recognize it after seeing the answer, that is an Again or Hard.
  • Early failures cost little. A lapse on a new card simply shortens its interval. It is the cheapest time to find out you did not know it.

Because the rating drives everything, Paivio schedules reviews from explicit recall ratings. You tell it how the recall went, and the next date follows from that.

SM-2: the multiplier method

SM-2 is the best-known spaced repetition algorithm. It was developed by Piotr Woźniak for SuperMemo, and Anki’s original scheduler was based on it. Anki’s own FAQ describes the two algorithms it offers: one derived from SuperMemo 2, and a newer one called FSRS, available from Anki 23.10.

The idea is simple:

  • Each card carries an ease factor, a number that represents how easy the card has been for you.
  • The first reviews use fixed short steps, on the order of a day and then several days.
  • After that, the next interval is the previous interval multiplied by the ease factor.
  • Passing grades nudge the ease factor up or hold it steady, and failing a card resets the interval and lowers the ease.

So a card you keep recalling well climbs quickly: days, then weeks, then months. A card you keep failing stays on short intervals.

Strengths: it is easy to understand, needs no training data and behaves predictably. You can look at a card and roughly predict its next interval.

Where it gets rough: the multiplier is a rule of thumb, not a model of your memory. Every card follows the same formula shape regardless of how you actually forget. Cards that you fail once and recover can end up with a lowered ease that is slow to recover, a pattern Anki users call “ease hell.” The system also has no direct dial for how much you want to remember, so you cannot say “aim for 90 percent” and let the schedule follow.

FSRS: scheduling from a memory model

FSRS, the Free Spaced Repetition Scheduler, takes a different approach. Instead of a fixed multiplier, it estimates the state of your memory for each card and predicts the chance you will recall it at any future date. The Anki FAQ linked above lists it as the second built-in algorithm. Three ideas describe it:

  • Stability: how long a memory lasts, roughly how many days until recall probability falls to a set level.
  • Difficulty: how hard the card is for you, based on your ratings.
  • Retrievability: the current probability that you would recall the card right now.

You choose a desired retention, such as 90 percent, and the algorithm sets each card’s next review for the day its predicted recall falls to that figure. This links directly to the optimal-schedule idea above, where recall probability itself drives when to review.

The practical difference is control. With a multiplier method you tune intervals indirectly. With a retention-targeted model you set the goal and the workload follows: a higher target means more reviews each day, a lower one means fewer reviews and more forgotten cards.

SM-2 and FSRS side by side

SM-2 style FSRS style
Core mechanism Previous interval times an ease factor Predicted recall probability from a memory model
Main user control Interval modifiers, steps Desired retention
Uses your review history Per-card ease only Per-card memory state, with parameters fitted to your history
Predictability Easy to reason about by hand Easier to reason about as a retention target
Best fit Small decks, simple routines Large, cumulative decks with long review histories

Neither is magic. Both are only as good as the ratings you feed them.

What the research says about results

The honest picture for exam prep is encouraging, with some care about how strong the evidence is.

Medical education. A 2026 systematic review of Anki use and academic performance found positive associations between regular use and USMLE Step 1 performance in three studies. High-frequency users outperformed minimal users by 4 to 13 points. Results for course-based exams were mixed, and the evidence was largely observational, meaning students who study more use more cards, and the data cannot fully separate cause from habit. The review’s reading is that the benefits fit established learning theory for long-term foundational knowledge, which is what boards test.

Successive relearning. A study of successive relearning, which combines retrieval practice with spaced practice through a flashcard program, reported that using it in courses improved performance on a high-stakes exam by at least 10 percent across experiments.

Personalized schedules. In a study of retrieval practice software, a personalized review scheduler produced 16.5 percent better course retention than massed study and 10.0 percent better than a one-size-fits-all spaced scheduler, measured on cumulative exam outcomes. This is the strongest argument for algorithms that adapt to the individual rather than apply one fixed pattern.

Interval design. A randomized trial in school science, SMART Spaces, found that a model with 24-hour gaps between repetitions improved attainment against a control group (effect size 0.19), and that student engagement predicted gains. Spacing helps, but only if students actually do the reviews.

The common thread: spacing plus retrieval plus consistency. If you are tuning for a standardized exam such as the MCAT, the same logic applies, and our MCAT flashcards strategy shows how to build and schedule the cards.

What the research says about results

Fixed schedules versus an algorithm

You will see simple fixed patterns online, such as reviewing on days 2, 3, 5 and 7 after first learning something. They are better than rereading once and hoping, and they require no software.

The limit is that a fixed calendar treats every fact the same. The pharmacology mnemonic you nailed and the renal physiology pathway you keep missing get the same dates. An algorithm lets the first fade out and the second come back sooner. For a cumulative subject with thousands of facts, that difference compounds. If you want to see concrete intervals, our page on the spaced repetition schedule for exams lays out a calendar you can start from.

Fixed review dates compared with adaptive intervals

Choosing settings for a personalized schedule

Whatever tool you use, five decisions shape your results.

1. Pick a retention target you can sustain

A target around 90 percent is a common starting point. Raising it increases daily reviews. Before a board exam, a slightly higher target is reasonable. During a heavy course week, a lower one keeps the queue manageable. Change it by small steps and watch the workload for a week before changing again.

2. Tie the schedule to your exam date

A schedule that ignores the calendar can leave you with cards due long after your exam. Exam-date pacing compresses or stretches intervals so that your material comes around the right number of times before test day. Students who take standardized tests outside medicine, such as those comparing the best ACT prep courses, face the same question of how to fit structured review into a fixed timeline.

3. Control new cards

The daily queue is mostly set by how many new cards you add. Adding 100 new cards on one day creates reviews that echo for weeks. Add them at a rate you can keep through your busiest weeks.

4. Keep cards atomic

One fact or one idea per card. A card that asks for a list of eight items gets rated Hard every time, and the algorithm cannot tell which item you forgot. Good cards make ratings meaningful. Our guides on custom flashcards versus premade decks and making flashcards from a PDF cover how to build them.

5. Rate honestly, every time

This is the setting nobody can change for you. If you are unsure whether you recalled something, treat it as a miss.

Algorithm versus active recall

The algorithm decides when. Active recall decides how you study at each review. You need both: scheduled reviews that are passive rereading do little, and active recall at the wrong times wastes effort. The difference is covered in detail in spaced repetition versus active recall.

Where Paivio fits

We built Paivio around a simple observation: the algorithm matters less than whether it runs on your own material and honest ratings.

Students in pre-med, medical and nursing programs already have material they trust: lecture notes, textbook chapters and Anki decks they have spent months building. Throwing that away to adopt a new tool is a bad trade. So we start from what you already have. You can import your Anki decks with review history, or bring notes and topics, and keep studying without starting over.

Scheduling then follows your explicit recall ratings, so the system learns from what you actually remember. We add concept mapping because facts in cumulative subjects are not isolated: seeing how a mechanism, a drug class and a lab finding connect makes each card easier to recall. The study loop of review, rate, reschedule and connect is the core of the product, and machine-learning enhancements sit on top of it.

Our view is that a good schedule should adapt to the learner, not the other way around, and that an exam date should shape the plan.

Jackson

Start with your own deck

Paivio imports your own Anki decks, notes or topics and builds a review system around them, with daily review queues, exam-date pacing, concept mapping and AI-generated cards. It schedules from your explicit recall ratings. You can see the full workflow on the How it works page, and the research background on the science page.

Paivio

The Free plan costs $0 per month. Plus is $20 per month or $200 per year, and Pro is $40 per month or $400 per year, with different credit and AI usage limits. Credit top-ups are $10 one-off. The platform is in beta, with unlimited cards. To bring a deck across, follow the guide on importing Anki decks, then try Paivio with the material you already have.

FAQ

What is the best spaced repetition algorithm?

For most students, the best algorithm is the one that matches your workload and that you will rate honestly. SM-2 style schedulers are simple and predictable. Retention-targeted models like FSRS suit large cumulative decks because you can set the recall level you want and let the intervals follow.

Is the SuperMemo algorithm still used?

Yes. SM-2 was developed by Piotr Woźniak for SuperMemo, and Anki’s original scheduler is based on it. Newer options such as FSRS have been added alongside it rather than replacing it entirely.

How often should I review a card?

The algorithm decides, based on your ratings. A new or failed card returns within a short interval, and a card you recall well moves out to longer gaps. Your job is to clear the daily queue and rate honestly. If the queue feels unmanageable, reduce new cards before changing anything else.

Does spaced repetition improve exam scores?

The evidence is positive but mostly observational. A systematic review in medical education linked regular Anki use to higher USMLE Step 1 performance, with results for course exams more mixed. Consistent daily review is what the benefit depends on.

Sources

Recommended

Your next chapter

Bring one deck.
Start one session.

Create an account and make your next review a place to begin.

Start studying free

Free during beta · No credit card · For students 18+