FSRS vs SM-2: Which Spaced Repetition Algorithm Is Better?

If you've set up a serious flashcard app, you've met a choice: FSRS or SM-2. Both decide when to show you each card, and both are far better than guessing. But one is a 1980s rule of thumb and the other is a memory model fitted to hundreds of millions of real reviews. Here's how they differ, what the evidence says, and which to pick.

The short answer

FSRS, for almost everyone. On the largest public benchmark — about 350 million Anki reviews from roughly 10,000 users — FSRS predicted recall more accurately than SM-2 for 99.6% of users. Better predictions mean fewer reviews wasted on cards you already know and fewer lapses on cards you were about to forget. SM-2 is still a sound, simple algorithm, and either one used daily beats no spaced repetition at all.

What they have in common

Both are spaced-repetition schedulers: after you review a card and say how well you knew it, they choose the next review date — pushing well-known cards further out and bringing struggling cards back sooner. The goal is the same: review each card right before you'd forget it, and no more often than necessary.

SM-2: the classic rule of thumb

SM-2 was written by Piotr Wozniak for SuperMemo between 1987 and 1989, and a modified version has been Anki's default scheduler ever since Anki appeared. Its core is one number per card, the ease factor:

• Every card starts with an ease of 2.5. The first intervals are fixed (1 day, then 6 days in the original).
• After that, each successful review multiplies the interval by the ease — a 10-day interval at ease 2.5 becomes 25 days.
• Rate a card badly and the interval collapses and the ease drops (never below 1.3). Rate it easy and the ease rises.

Strengths: simple, transparent, predictable, and battle-tested for decades.

Limits: it doesn't model your memory — it applies the same multipliers to everyone. And it has a well-known failure mode, ease hell: every lapse lowers a card's ease, easy ratings are rare, so difficult cards drift down to the 1.3 floor and keep coming back at short intervals long after you've learned them.

FSRS: a model of your memory

FSRS (Free Spaced Repetition Scheduler) was created by Jarrett Ye, building on his published research at the language-learning company MaiMemo, and is open source. Instead of one ease number, it tracks three things for every card:

• Stability — how long the memory lasts: the number of days until your chance of recalling it falls to 90%.
• Difficulty — how hard this particular card is for you.
• Retrievability — your probability of recalling it right now.

Each rating updates the card's stability and difficulty. FSRS then schedules the next review for the day your predicted recall falls to a desired retention you choose — 90% by default. Raise it and you get shorter intervals and more reviews; lower it and you get fewer reviews for slightly more forgetting. (What to set it to, with a year-long simulation.) The Anki manual warns that above 90% the workload climbs quickly, and above 97% it can be overwhelming.

FSRS has shipped inside Anki since version 23.10 (late 2023) as an opt-in option. It isn't switched on by default — you enable it in deck options.

What the benchmark found

The open-spaced-repetition project maintains a public benchmark that asks each algorithm to predict, before every review, the probability that you'll remember the card — then scores those predictions against what actually happened. On about 350 million reviews from roughly 10,000 Anki collections:

FSRS-6 had lower error than SM-2 for 99.6% of users. An AUC of 0.54 is barely better than a coin flip at telling which cards you're about to forget.

Two honest caveats. First, this measures prediction accuracy, not workload directly — SM-2 was never designed to output probabilities, so the benchmark's own authors note that no comparison between them is perfectly fair. Second, "fewer reviews" depends on your settings: the saving comes from FSRS spending reviews where recall is actually dropping instead of on a fixed multiplier, but set desired retention to 97% and FSRS will happily give you more reviews than SM-2.

Side by side

Should you switch?

• Starting fresh? Use FSRS with desired retention at 90%. There's nothing to migrate and the defaults work from day one.
• Years of SM-2 history in Anki? Switching is safe — FSRS reads your existing review history. Turn it on in deck options, keep retention at 90% at first, and use Anki's optimise button to fit the parameters to your own reviews.
• Happy with SM-2 and hitting your targets? There's no emergency. The habit matters far more than the algorithm; switch when you next have a quiet week.
• Drowning in reviews of cards you know well? That's classic ease hell, and FSRS is the fix.

For developers

FSRS isn't tied to Anki. The open-spaced-repetition organisation publishes implementations in several languages, including Rust (fsrs-rs, which Anki uses), Python (py-fsrs), TypeScript (ts-fsrs) and Go (go-fsrs), so any app can schedule with it.

FSRS and SM-2 in StudyTab

StudyTab schedules with FSRS-6 (via the open-source ts-fsrs library) and keeps SM-2 available, configurable per deck. It uses the default FSRS-6 parameters — fitted on the large Anki dataset above — and updates every card's stability and difficulty from your ratings. Retention presets let you run core subjects at 90% and breadth material at 85%, and the schedule calendar shows the reviews due over the next 30 days. Importing from Anki? Your decks keep their due dates and intervals.

Bottom line

SM-2 is the reliable classic; FSRS is the more accurate successor. On 350 million real reviews it predicted recall better for almost every user, and it avoids ease hell. Pick FSRS, set desired retention to 90%, then stop tweaking and start reviewing — the algorithm is a rounding error compared to showing up every day.

Want spaced repetition without the setup? Generate a deck and let the schedule run.