SM-2 vs. FSRS: How Anki's Scheduling Algorithms Work
The SM-2 vs FSRS question is about which of Anki's two scheduling algorithms decides when each card comes back for review. Anki includes both. The one your deck uses directly shapes how many reviews a Japanese learner faces every day.1
Overview
A scheduler is the engine that picks each card's next review date after you grade it. Anki has two: SM-2, its legacy default derived from SuperMemo 2, and FSRS, the Free Spaced Repetition Scheduler.1
The two algorithms in one sentence each
SM-2 is a fixed arithmetic formula that scales each card's next interval by a per-card "ease factor." It is calculated without an underlying model of how memory works.2
FSRS is a memory model whose parameters are trained on your review-log data. It predicts your probability of recall and schedules from that prediction rather than from a fixed multiplier.34
The name "FSRS" expands to Free Spaced Repetition Scheduler. Its optimizer, benchmark, and reference implementation are all open source under the open-spaced-repetition organization.456
Which one is your deck on right now
FSRS support was added to Anki in version 23.10, released on 2023-10-31. The release notes say: "Support for FSRS (which improves upon the scheduling provided by SM-2) is now integrated into Anki. You can compute the model weights directly inside Anki, and no longer need to use custom scheduling."7
FSRS has been opt-in since Anki 23.10 (October 2023). The Anki FAQ describes the situation at that release this way: "As of Anki 23.10, Anki has two available algorithms. The first one is based on the SuperMemo 2 algorithm." In other words, FSRS is the alternative the user enables, not an automatic replacement.17
The manual states that "FSRS can only be enabled globally; you cannot enable it for some presets and disable it for others." Enabling it is a single user-controlled setting that applies to your whole collection.8
A maintainer opened a proposal on 2024-12-06 (issue #3616) to make FSRS the default out of the box for new users: "In the next non-trivial (not 24.11.x) update, I think it's about time we enable FSRS out of the box. Any objections?" That proposal remains open, with no closing release pinned, so no specific default-on version can be stated here.9
The practical consequence is simple: a deck or collection created before you enabled FSRS, or one used on an un-updated client, may still be running SM-2.18
How to check
Deck Options shows the active algorithm. FSRS appears there as the toggle described in the manual's Deck Options section.8 The click-by-click path belongs in the dedicated Anki tool walkthrough rather than here.
How SM-2 works
SM-2 is the second SuperMemo algorithm, specified by Piotr Wozniak in 1990. It predates Anki by decades and is the ancestor of the ease-factor schedulers that most flashcard apps still default to.2
The ease factor and the interval formula
SM-2 stores one state variable per item, the E-Factor (EF). It starts at 2.5 and has a floor of 1.3.2 The interval schedule scales off that single number:
I(1) := 1
I(2) := 6
for n > 2: I(n) := I(n-1) * EF
Here I(n) is the inter-repetition interval in days. After the first two fixed steps, each new interval is simply the previous interval multiplied by the card's ease.2
The E-Factor itself is updated after every graded review by the following formula, where q is the response quality on a 0–5 scale:
EF' := EF + (0.1 - (5 - q) * (0.08 + (5 - q) * 0.02))
After the update, EF' is clamped to a minimum of 1.3.2 The original SM-2 quality scale runs from "5 - perfect response" down to "0 - complete blackout." It includes grades such as "2 - incorrect response; where the correct one seemed easy to recall."2
When response quality falls below 3, SM-2's lapse rule restarts the repetition count from I(1) without changing the EF in that same step.2
Anki's implementation differs from the 1990 spec in documented ways: customizable initial learning steps, four answer buttons (Again/Hard/Good/Easy) rather than six numeric grades, and added protection against cards getting stuck in "low interval hell."1
The per-card multiplier that Anki exposes as "ease" is this E-Factor. The next interval is essentially the current interval times ease. That makes the schedule multiplicative and formula-driven, with no estimate of how likely you are to recall the card.21
Ease hell: why SM-2 over-reviews cards you know
The E-Factor drops whenever you lapse, and the SM-2 formula has no mechanism to raise it back once it has fallen.2 With the 1.3 floor in place, a card that has lapsed several times can stay pinned near minimum ease.
Intervals for such a card then grow only slowly, multiplied by a number near 1.3 instead of 2.5 or higher. A card you now know well keeps resurfacing at short intervals. The learner community calls this failure mode "ease hell."24
The diagram below contrasts how a single lapse propagates under each algorithm. This is the cleanest way to see why SM-2 traps a card and FSRS does not.
How FSRS works
FSRS replaces SM-2's single ease number with a three-component model of memory, fit to your review history by machine learning.1 Instead of one multiplier, it uses three interacting variables that together estimate when a card is about to be forgotten.
The three variables: Difficulty, Stability, Retrievability
FSRS tracks retrievability, stability, and difficulty, instead of SM-2's single ease number.1 The ABC of FSRS defines each one precisely:
| Variable | Definition |
|---|---|
| Retrievability (R) | "The probability that the person can successfully recall a particular piece of information at a given moment."4 |
| Stability (S) | "The time, in days, required for R to decrease from 100% to 90%."4 |
| Difficulty (D) | "The inherent complexity of a particular information. It represents how difficult it is to increase memory stability after a review."4 In plain terms, harder cards gain stability more slowly. |
Because the model is fit to the whole review history, the next interval reflects a card's entire past, not only the last button you pressed. That is the structural difference from SM-2's last-grade-only update.34
Difficulty is also the structural fix for ease hell. Unlike SM-2's ease, which only ever decreases, FSRS's Difficulty is bounded and can move back toward the mean. A card therefore does not get permanently pinned at high difficulty and short intervals.4
Desired retention: the dial SM-2 never had
FSRS exposes a desired retention setting: a target probability of recall that the scheduler aims to hit when a card comes up for review.8 SM-2 has no equivalent. Its schedule is fixed by the EF formula and gives the learner no target-recall dial to turn.28
The default is 90%. The manual says this schedules "cards so you have a 90% chance of remembering them when they come up for review again" and notes that the default "offers a good balance of retention and workload."8
Pushing the dial higher costs reviews quickly. The manual recommends keeping desired retention below 97%, warning that "Above 90% the workload increases very quickly, and above 97% the workload can be overwhelming."8
The optimizer trains FSRS parameters on your own review log, not on someone else's averages. The manual notes that "the optimizer requires several reviews to fine-tune the parameters." It also flags a "Low number of reviews (less than a few hundred)" as a health-check limitation, without naming a hard threshold.8
What "20-30% fewer reviews" actually means
The headline figure comes from the project itself, and the source labels it carefully. The ABC of FSRS states: "With FSRS, users have to do 20–30% fewer reviews than with SM-2 algorithm to achieve the same retention level."4
That parenthetical matters. The number is a simulation result reported by the open-spaced-repetition project, not an independently measured guarantee. It is best read as a tendency rather than a promise.4
The Anki FAQ states the same direction more cautiously: "With FSRS, users have to do fewer reviews than with Anki's default algorithm to achieve the same retention level."1 In the issue #3616 discussion, the maintainers cite the open-spaced-repetition benchmark. That benchmark reports FSRS with default parameters as "better than SM-2 in 91.9% cases," improving further with optimization. This is the benchmark's claim rather than an independent finding.59
"Fewer reviews" is about total volume at equal retention, not uniformly longer gaps everywhere. The FAQ notes that "FSRS tends to give longer first intervals than SM-2, but for mature cards the opposite is true - FSRS is more conservative."3
For a Japanese deck, the practical payoff is that the same cards should generate fewer total reviews per day under FSRS than under SM-2 at the same target retention, per the project's simulation. Treat it as a tendency that lightens daily load, not a fixed dividend.14
Should you switch, and when
The default recommendation
For essentially all learners on a current Anki version, enable FSRS. The benchmark results and the algorithm's design both favor it over SM-2 at equal retention. The desired-retention dial also gives you direct control that SM-2 cannot offer.145
The cases where staying on SM-2 is tolerable are narrow. A very small or very young deck gives the FSRS optimizer too little review history to fine-tune, per the "low number of reviews" health-check limitation. A setup forced to sync with an old client that predates FSRS support also cannot rely on the scheduler being present.18
For a learner who wants FSRS's efficiency without hand-building and maintaining an Anki deck, Amenokori is built around the FSRS algorithm out of the box. Its scheduler is already the one this article recommends, with no SM-2-to-FSRS migration or optimizer setup to run yourself.10
What switching does and does not change
Enabling FSRS does not delete a card's history or reset your progress; it reschedules existing cards based on the history they already carry.83
There is one data nuance worth knowing. When FSRS reschedules existing cards, the manual notes that the action "will add a review entry to each of your cards." This grows collection size even though your progress and history are preserved.8
The first optimize run needs some accumulated review history to be useful, in line with the optimizer's "requires several reviews" limitation.8 The exact click path for changing the scheduler, running the optimizer, and setting desired retention belongs in the dedicated Anki tool walkthrough rather than this conceptual overview.
Good to know
SM-2's "2" is the second SuperMemo algorithm
The canonical SM-2 specification is Piotr Wozniak's, published in 1990 as Algorithm SM-2. It is the SuperMemo-2 method and the ancestor of the ease-factor schedulers that flashcard apps still ship as defaults.2
"FSRS" expands to Free Spaced Repetition Scheduler
The name is literal. Its optimizer, benchmark, and reference implementation are open source under the open-spaced-repetition organization, in the fsrs4anki, srs-benchmark, and awesome-fsrs repositories.456
Setting desired retention to 95%+ "to be safe" balloons daily reviews
Raising desired retention well above the default in the hope of feeling safer is the wrong move. It multiplies your review load without a matching benefit. Keep the default 90% unless you have a specific reason to change it. The manual warns that "Above 90% the workload increases very quickly, and above 97% the workload can be overwhelming" and recommends staying below 97%.8
Expecting an instant payoff from FSRS
FSRS does not earn its headline efficiency on day one. The optimizer fits parameters to your own log and "requires several reviews to fine-tune." With a low number of reviews, fewer than a few hundred, it cannot yet schedule at full strength.8
Assuming every mobile or sync client has FSRS
FSRS was integrated into Anki at version 23.10 in 2023, and the scheduler is a collection-level setting. Very old AnkiDroid or AnkiMobile builds predate it, so an un-updated client can still behave as if it were running SM-2.187
See also
- Spaced Repetition and the Forgetting Curve: Why Reviewing on a Schedule Works
- When SRS Becomes Counterproductive: Anki Burnout, Leeches, and the Exit Signs
- Why Anki Has Become Painful (and How to Fix It): A Diagnostic Triage
- Beyond Anki: SRS Tools and Approaches Compared
- Yomitan + Anki: One-Click Card Creation