# FSRS vs SM-2: Which Spaced Repetition Algorithm Wins in 2026?

URL: https://fulinlabs.com/blog/fsrs-vs-sm2-spaced-repetition-2026/
Published: 2026-07-17
Updated: 2026-10-02
Product: flashcard
Summary: FSRS vs SM-2 compared plainly: how each algorithm models memory, which one schedules fewer reviews at equal retention, and which to actually pick in 2026.

I'll say the biased part up front: I built [Flashcard](https://flashcard.fulinlabs.com), a spaced-repetition tool, so I have a stake in this conversation. But the FSRS vs SM-2 question is one I get asked honestly enough, by people comparing algorithms before they've even picked an app, that it's worth answering straight — including the parts where the answer isn't "use my tool."

**Anki supports both FSRS and its legacy SM-2 scheduler. FSRS is available in Anki 23.10 and
later; check Deck Options to see whether it is enabled in your collection.** The
[Anki manual](https://docs.ankiweb.net/deck-options.html#fsrs), checked October 2, 2026,
describes FSRS as an alternative to SM-2 and explains how to turn it on. Availability and a
default setting are different claims.

## FSRS vs SM-2 at a glance

| | SM-2 | FSRS |
|---|---|---|
| Year / origin | 1987, SuperMemo (Piotr Woźniak) | Open-source scheduler; integrated support in Anki 23.10+ |
| How it models memory | One "ease factor" per card, adjusted up or down after each review | Three separate variables per card: difficulty, stability, and retrievability |
| Review efficiency | Rule-based intervals adjusted through an ease factor | Uses a memory model and desired retention; Anki says it can improve recall for the same study time |
| Data needed to work well | Works reasonably from the first review | Improves as it accumulates review history; needs a real usage history to fit parameters well |
| Best for | Tools without FSRS support or an established workflow you prefer to keep | Learners using compatible clients who want scheduling based on desired retention |

## What is SM-2?

SM-2 is the scheduling algorithm SuperMemo introduced in 1987, and it's the one nearly every spaced-repetition app you've used — Anki for most of its life included — was built on. It works by giving every card a single "ease factor" that goes up when you recall the card easily and down when you struggle, and it uses that one number to decide how much to stretch the interval before the next review.

That simplicity is SM-2's strength and its limit at the same time. Because it collapses "how hard is this card" and "how well do I currently remember it" into one adjustable number, it can't distinguish a card that's inherently difficult from one you just haven't reviewed in a while. The practical effect is that SM-2 tends to be a little conservative — it shows you cards more often than strictly necessary to hit your retention target, because it doesn't have the resolution to know it could safely wait longer.

SM-2's other defining trait is that it's stateless in a useful way: it doesn't need a large history of your review data to start making reasonable scheduling decisions. A brand-new deck with a handful of reviews logged behaves about as sensibly under SM-2 as one with years of history. That's part of why it spread to so many flashcard apps beyond SuperMemo itself — it's simple to implement, predictable to reason about, and doesn't require any machine-learning-style fitting step to work. For nearly three decades, "spaced repetition" and "SM-2" were more or less synonymous in the flashcard-app world.

## What is FSRS?

FSRS (Free Spaced Repetition Scheduler) is a modern, open-source scheduling algorithm that models each card's difficulty, stability, and retrievability as three separate values instead of one blended ease factor. Difficulty captures how inherently hard the card is for you; stability captures how long you can currently go between reviews before forgetting; retrievability is the live estimate of your recall probability at any given moment.

Because those three things are tracked independently, FSRS can be more precise about when a card actually needs to resurface. A card that's easy but hasn't been reviewed in a while gets treated differently than a card that's genuinely hard, even if both currently sit at similar recall probability — a distinction SM-2 can't make. This is the reason FSRS schedules fewer reviews for the same retention target: it's not being more aggressive, it's being more accurate about which cards are safe to leave alone longer.

Personalization improves with review history. Anki's optimizer fits parameters to reviews in
the selected preset; it is not a separate model fitted independently for every card. The
[FSRS parameters documentation](https://docs.ankiweb.net/deck-options.html#fsrs-parameters)
notes that a low number of reviews, fewer than a few hundred, can affect optimizer performance.
That does not establish a 1,000-review threshold for using FSRS or prove that a beginner should
choose SM-2. Do not copy someone else's parameters or edit them manually.

## So which one actually wins?

FSRS is a reasonable choice if your clients support it and you want to choose a desired
retention target. Anki's manual says its more accurate model of forgetting can help you remember
more material in the same amount of time. That supports considering FSRS; it does not promise
a fixed percentage reduction in your reviews. Your cards, grading habits and chosen retention
all affect the result.

That said, SM-2 is genuinely fine in a couple of specific situations:

- **One of your clients does not support FSRS.** Check every device you use before switching;
  the manual lists Anki/AnkiMobile 23.10+, AnkiWeb, and AnkiDroid 2.17+ as compatible.
- **You're cramming for something short-term**, like an exam in two weeks, where long-run scheduling efficiency doesn't matter because you're not trying to retain the material for years.
- **Your current tool doesn't support FSRS** and switching tools just to get a marginally better scheduler isn't worth the disruption to a habit that's already working.

Before switching, read the manual's setup guidance and check your learning steps and grading
habits. In particular, **Again** means forgotten; **Hard** means recalled with difficulty.
Using Hard for a forgotten answer gives the memory model the wrong information.

Leave **Reschedule cards on change** off when first switching unless you understand the effect:
the [manual](https://docs.ankiweb.net/deck-options.html#reschedule-cards-on-change) warns that
rescheduling can make many cards due immediately. Without it, future reviews use the new
scheduler without an immediate change to all due dates.

## What retention target should I set?

Anki's documented default [desired retention](https://docs.ankiweb.net/deck-options.html#desired-retention)
is **90%**. It is the probability the scheduler aims for when a card becomes due, not a guarantee
that you will remember that fraction on every day. Raising it increases workload; lowering it
allows more forgetting in exchange for fewer reviews. Anki's **Help Me Decide** simulator can
estimate that tradeoff for your collection.

A few practical notes on choosing a number:

- **Start from the default**, then check whether its review workload fits the time you have.
- **Compare targets in the simulator** before assuming a higher number is affordable.
- **If reviews become unsustainable**, inspect new-card intake, overdue cards and retention
  together. Queue growth is not caused by retention alone.

There's no universally "correct" retention value — it's a genuine tradeoff between review time and forgetting rate, and the right setting depends on what you're actually optimizing for.

For a transparent general model, the [flashcard workload calculator](/tools/flashcard-workload-calculator/)
compares 85%, 90%, and 95% at a chosen new-card rate. It is an illustrative workload model,
not an FSRS simulation or a prediction fitted to your Anki history; use Anki's simulator for that.

## The part where I'm honest about my own bias

Here's where the algorithm question actually connects to the tool question, and where I want to be careful not to overclaim. Flashcard doesn't advertise FSRS. It uses spaced repetition scheduling, full stop — I'm not going to tell you it runs a specific named algorithm it doesn't, because that's exactly the kind of claim this article is arguing against making carelessly.

My design priority for Flashcard is making decks quick to build and review. A scheduling model
can help, but it cannot do the reviews for you. That is a product-design preference, not evidence
that one app or algorithm outperforms another for every learner.

If you're deciding between apps on the basis of which scheduler they run, that's a reasonable thing to research — this article exists because it's a legitimate question. But if you're deciding between "the app with the better algorithm that I'll open twice a month" and "the app I'll actually open every day," pick the second one, whatever algorithm sits under the hood.

## Related reading

If you want the fuller mechanics of how spaced-repetition scheduling works in general, I go through it in more plain-language detail in [how spaced repetition scheduling works](/blog/spaced-repetition-explained/). If you're studying Chinese specifically and want to see this applied to a real workflow, [spaced repetition for Chinese SRS](/blog/spaced-repetition-chinese-srs-2026/) covers that ground. And if you're weighing Flashcard against Anki directly rather than just the algorithms underneath them, [Flashcard vs Anki for Chinese vocabulary](/blog/flashcard-vs-anki-chinese-2026/) is the honest side-by-side.

The other question that tends to come right after "which algorithm is better" is "how many cards should I actually be reviewing per day" — that's a separate, equally important variable, and I cover it in [how many flashcards to study per day](/blog/how-many-flashcards-per-day-2026/).

You can try Flashcard's spaced-repetition decks at [flashcard.fulinlabs.com](https://flashcard.fulinlabs.com) — no algorithm marketing, just a tool built to be fast enough that you'll actually use it tomorrow.

## FAQ

### Is FSRS better than SM-2?

For most learners, yes. FSRS models memory with three separate variables (difficulty, stability, retrievability) instead of SM-2's single ease factor, which lets it schedule noticeably fewer reviews at the same retention target. SM-2 is still perfectly workable, especially for small decks or short study windows, but FSRS is the better default in 2026.

### What spaced repetition algorithm does Anki use?

Anki supports FSRS in version 23.10 and later, alongside its legacy SM-2 scheduler. The manual explains how to enable FSRS in Deck Options. Support in 23.10 does not mean it became the default in that version; check your own deck options rather than assuming it is enabled.

### What should I set my FSRS retention to?

Anki's documented default desired retention is 90%. Higher retention increases review workload; lower retention reduces it while allowing more forgetting. Use Anki's workload simulator to choose a sustainable target rather than treating one value as right for everyone.

### Can a beginner use FSRS with little review history?

Yes. A small review history limits how well FSRS can personalize its parameters, but it is not a reason to claim that FSRS cannot be used. Anki's manual notes that fewer than a few hundred reviews can affect optimizer performance; it does not prescribe a 1,000-review minimum for using FSRS.

### Does Fulin Flashcard use FSRS or SM-2?

Flashcard uses spaced repetition scheduling rather than committing to either named algorithm as a branded feature. The design goal is a tool fast enough to build decks in and open daily, on the view that consistent daily use matters more than which specific scheduler is running underneath.