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

The short answer

FSRS predicts more accurately when you’ll forget a card, so it can schedule fewer reviews for the same retention. In the largest public benchmark, even FSRS with its default parameters had about half of SM-2’s prediction error.

SM-2 still works. It’s simpler, its intervals are easy to predict, and learners have used it for decades. If your reviews feel manageable, you don’t have to switch.

A note on bias: I make Azri, which uses FSRS. Every number below links to its source so you can check it.

FSRS vs SM-2 at a glance

SM-2 sets intervals with a rule of thumb. FSRS estimates how likely you are to remember, then picks the interval that matches your target.

SM-2 as described by Piotr Woźniak and in Anki’s FAQ; FSRS as described by the open-spaced-repetition project. Benchmark figures are explained below.
Feature	SM-2	FSRS
Created	1987, by Piotr Woźniak for the first SuperMemo program.	From 2022, by Jarrett Ye and the open-spaced-repetition community. FSRS-5 arrived in 2024.
What it tracks per card	An ease factor, starting at 2.5, and the last interval.	Difficulty, stability and retrievability.
How it sets the next interval	1 day, then 6 days, then the last interval × the ease factor.	The day your predicted chance of recall falls to your target retention.
Predicts your chance of remembering	No.	Yes.
Late reviews	The original ignores how late you are. Anki’s version takes it into account.	Uses the real time since your last review.
Main setting	In Anki: starting ease, learning steps, interval modifier and more.	Desired retention, 90% by default.
Fitting to you	Each card’s ease drifts with your answers.	Optional: Anki can fit 19 to 21 parameters to your review history.
Benchmark log loss (lower is better)	0.72, or 0.62 for Anki’s version.	0.382 with default parameters, 0.356 when fitted to each user.
Where you’ll find it	Anki (a modified version, unless you switch on FSRS) and Mnemosyne.	Anki (since 23.10), Azri and other apps.

How SM-2 works

Piotr Woźniak wrote SM-2 for the first SuperMemo program in 1987 and published it in 1990. It’s short enough to run by hand.

After each review you grade your answer from 0 (complete blackout) to 5 (perfect).
The first interval is 1 day, the second 6 days.
After that, each interval is the previous one multiplied by the card’s ease factor, which starts at 2.5.
Each grade nudges the ease factor up or down, but never below 1.3. A grade below 3 restarts the card’s intervals and keeps its ease factor.

Anki’s version is a modified SM-2. It has four buttons instead of six grades, adds learning steps for new cards, and changes ease in fixed steps: Again lowers it by 20 percentage points, Hard by 15, Easy raises it by 15, and Good leaves it alone (Anki FAQ). Since Good never raises ease, a card you struggled with early can stay on short intervals for a long time.

What SM-2 doesn’t do is estimate how likely you are to remember a card. Its intervals are a rule of thumb, which is also why they’re easy to understand.

How FSRS works

FSRS (Free Spaced Repetition Scheduler) was created by Jarrett Ye, building on memory research he published with the language-learning app MaiMemo (KDD 2022). It’s developed in the open by the open-spaced-repetition community.

Each card has a difficulty, a stability (how many days until your chance of recall drops to 90%) and a retrievability (your chance of recalling it right now).
After each Again, Hard, Good or Easy, FSRS updates difficulty and stability with formulas whose parameters were fitted to real review histories.
It schedules the next review for the day retrievability is predicted to fall to your desired retention. Anki and Azri both default to 90%.
In Anki you can fit the parameters to your own history with the Optimize button (Anki manual). Apps without an optimizer, including Azri, use the defaults.

The result is one setting that means something: at 90% desired retention, FSRS aims for you to remember about 9 in 10 cards when they come due. Push it higher and you get more reviews; drop it and you get fewer, with more misses.

What the benchmark says

The open-spaced-repetition project runs the largest public comparison of scheduling algorithms. It replays real Anki review histories and checks how well each algorithm predicted whether the person would remember the card.

Unweighted averages over 9,999 Anki collections (349,923,850 reviews). Source: srs-benchmark README, 23 April 2025, the last version that listed both FSRS-5 with default parameters and SM-2.
Algorithm	Log loss ↓	RMSE (bins) ↓	AUC ↑
FSRS-5 fitted to each user	0.356	0.074	0.701
FSRS-5 default parameters, as in Azri	0.382	0.102	0.693
Anki’s SM-2 default settings	0.62	0.172	0.613
SM-2 original	0.72	0.203	0.603

Log loss and RMSE (bins) measure calibration: whether a predicted 80% chance really means about 8 in 10 cards remembered. AUC measures how well an algorithm tells the cards you’ll remember from the ones you’ll forget.

With default parameters, FSRS-5’s RMSE is about half of SM-2’s (0.102 vs 0.203) and about 40% lower than Anki’s SM-2 (0.172). Fitting FSRS to each user helps further, mostly in calibration (0.102 to 0.074). The benchmark also reports that a fitted FSRS-5 variant estimated recall better than Anki’s default SM-2 for 99.4% of collections.

SM-2 wasn’t built to predict

SM-2 only outputs intervals. The benchmark adds formulas to turn them into probabilities, and its authors point out that this is the only reason SM-2 can be scored at all.

Prediction isn’t the same as grades

The benchmark measures prediction, not exam results. Better predictions let a scheduler hit your target with fewer wasted reviews: Anki’s FAQ says FSRS needs fewer reviews for the same retention.

It’s all Anki data

Every collection comes from an Anki user, so the results describe people studying Anki-style flashcards.

The benchmark has moved on

The current benchmark ranks FSRS-6, FSRS-7 and neural models, and no longer lists SM-2, which is why the table uses the April 2025 results. In the current results, FSRS-6 scores 0.346 log loss against FSRS-5’s 0.356.

When SM-2 is still a fine choice

A better predictor doesn’t make SM-2 useless. It depends on what bothers you about your reviews.

SM-2 is fine if

Your daily reviews feel manageable and you remember what you need.
You like intervals you can predict and explain.
You study on paper or in a spreadsheet, where a simple rule is a feature.
You’d rather not change settings you’ve tuned for years.

FSRS is worth it if

Your review pile feels heavier than your retention justifies.
You want to choose a retention target directly.
You often come back to cards after long breaks.
Some cards seem stuck on short intervals no matter what you press.

How Azri uses FSRS

Azri schedules every review with spaced repetition (FSRS), version 5, and it’s free on iPhone and iPad. The web app comes with Pro and uses the same scheduler and presets. There’s no SM-2 mode. You choose Default (90%), Aggressive (95%), Relaxed (85%) or your own target, and Azri uses FSRS’s default parameters, without a personal optimizer.

See spaced repetition (FSRS) in Azri

Questions about FSRS and SM-2

Is FSRS better than SM-2?

At predicting when you’ll forget a card, yes. In the open-spaced-repetition benchmark of 9,999 Anki collections, FSRS-5 with default parameters had about half of SM-2’s prediction error. Anki’s documentation says FSRS needs fewer reviews for the same retention. SM-2 still works, and it’s simpler to understand.

Does FSRS need a lot of review history?

No. FSRS starts with default parameters fitted to other learners’ reviews, so it works from your first review. In Anki, the Optimize button can later fit the parameters to your own reviews.

What’s the difference between FSRS-5 and FSRS-6?

FSRS-6 handles same-day reviews better and adds a parameter for the shape of the forgetting curve, so the curve can differ from person to person. In the benchmark, FSRS-6’s log loss is 0.346 against FSRS-5’s 0.356, a small step next to the one from SM-2 (0.72) to FSRS-5. The benchmark now also lists FSRS-7, the newest version.

What desired retention should I use with FSRS?

90% is the default in both Anki and Azri. Anki’s manual notes that the workload rises quickly above 90% and can become overwhelming above 97%.

Does Anki use FSRS or SM-2?

Both. Anki schedules with its own version of SM-2 unless you turn on FSRS in the deck options, which Anki has supported since version 23.10.

Can I use FSRS on iPhone for free?

Yes. Azri is free on iPhone and iPad and schedules every review with FSRS. AnkiMobile, Anki’s own iPhone app, also supports FSRS and costs $24.99 on the US App Store (checked 19 October 2026). See how Azri compares with AnkiMobile.