Skip to content

The Anki algorithm

Anki schedules reviews with one of two algorithms: its own variant of SM-2, the SuperMemo algorithm it has used since its start, or FSRS, built in since Anki 23.10. The SM-2-based scheduler is still the default; FSRS is an option you switch on.

Updated 5 min read

In short

  • Anki's default scheduler is based on SM-2: each time you answer Good, the interval is multiplied by the card's ease, which starts at 250%.
  • Anki changes SM-2 in several ways: four buttons instead of six grades, learning steps in minutes, and settings such as the easy bonus and interval modifier.
  • FSRS, available since Anki 23.10, predicts how likely you are to recall each card and schedules it when that probability falls to your desired retention (90% by default).
  • FSRS predicts recall more accurately in benchmarks; SM-2 is simpler to understand and predict. You can switch at any time.

What algorithm does Anki use?

Anki uses a scheduler based on SM-2, the algorithm Piotr Woźniak published for SuperMemo, and since version 23.10 it also offers FSRS (Free Spaced Repetition Scheduler).[1] Anki 23.10 came out in October 2023 and was the first release with FSRS integrated, instead of as a custom script.[2]

FSRS is not the default. As of Anki 26.09, the latest release in November 2026, the manual presents FSRS as an alternative to Anki's “legacy” SM-2 algorithm that you enable in the deck options, and it applies to your whole collection, not one deck.[3] In Anki's source code, the FSRS setting is off unless you turn it on.[4]

How is Anki’s scheduler different from SM-2?

Anki keeps SM-2's core idea, multiplying each interval by an ease factor, but changes how you grade cards and how ease moves. In Woźniak's SM-2, you grade each answer from 0 to 5, the first two intervals are fixed at 1 and 6 days, and a failed card restarts from the beginning.[5] Anki's defaults are below.[1][3][4]

SettingWoźniak’s SM-2Anki (default)
Grades0 to 5; three failing gradesAgain, Hard, Good, Easy; one failing button
First intervals1 day, then 6 daysLearning steps of 1 and 10 minutes, then 1 day (Good) or 4 days (Easy)
Starting ease2.5250%
Ease changesFormula applied after every answerAgain −20, Hard −15, Good 0, Easy +15 percentage points
Minimum ease1.3130%
Hard—Interval × 1.2 (hard interval)
Easy—Interval × ease × 1.3 (easy bonus)
Interval modifier—× 1.00, applied to every review interval
Forgotten cardRepetitions restartRelearning step of 10 minutes; new interval 0% of the old one, so at least 1 day
Random fuzzNoYes, cannot be turned off

Two smaller differences matter in practice. Anki rewards late reviews: if you answer correctly after the due date, the extra days are taken into account in the next interval.[1] And Anki adds a little random “fuzz” to each review interval so that cards added together do not stay due on the same day.[6]

How does Anki schedule a new card?

A new card first goes through short learning steps, then graduates to day-based intervals that grow with its ease. With the default settings:

  1. Learning. The card is shown again after 1 minute, then after 10 minutes, each time you press Good.[4] Again sends the card back to the first step; Easy skips the remaining steps.[1][3]
  2. Graduation. Good on the last step gives the first review interval: 1 day. Easy gives 4 days.[3][4]
  3. Review. Each Good multiplies the interval by the ease. At 250% the card comes back after roughly 1, 3, 8 and 20 days, with some fuzz.
  4. Lapse. Again on a review card lowers the ease by 20 points and sends the card through a 10-minute relearning step. Its next interval is then 1 day, the default minimum.[1][3]

Mistakes during learning do not lower the ease, so a card that was hard to learn starts its reviews at 250% like any other.[1] For a fuller picture of how intervals grow, see the spaced repetition schedule.

What is “ease hell”?

“Ease hell” is what Anki users call cards stuck at a low ease, often the 130% minimum, which keep coming back after short intervals even once you know them well. It is a community term, not one used in the Anki manual, and it is discussed on the Anki forums.[7]

It follows from how ease moves in Anki's SM-2 scheduler. Again and Hard lower the ease, Good leaves it unchanged, and only Easy raises it.[1] A card that you failed a few times keeps its low ease even after you start answering Good every time, so its intervals grow slowly. FSRS handles this differently: a card's difficulty drifts back toward a default value each time you answer Good, which the FSRS FAQ quoted in that thread says avoids the problem.[7]

How do you use FSRS in Anki?

You turn FSRS on in the FSRS section at the bottom of the deck options, then choose a desired retention: the share of reviews you want to answer correctly.[3] FSRS estimates each card's probability of recall and schedules the review for when it falls to that target. The default is 90%; a higher target means more reviews.[1][3]

  • Optimize. The “Optimize” button under “FSRS parameters” fits the model to your own review history. It needs some data: the manual warns that results are poor with fewer than a few hundred reviews.[3]
  • SM-2 settings. Options such as the graduating interval and easy bonus are hidden, and ease no longer drives intervals. The manual recommends learning and relearning steps shorter than 1 day.[3]
  • Existing cards. Your past reviews are kept and used. By default, due dates do not change: each card is rescheduled by FSRS the next time you review it. The “Reschedule cards on change” option moves them all at once, which the manual does not recommend when you first switch.[3]

For how FSRS models memory with difficulty, stability and retrievability, see the FSRS algorithm. For the app itself, see our Anki profile.

Should you use FSRS or SM-2 in Anki?

For most people FSRS is the better choice, because it predicts when you will forget more accurately; SM-2 is still a reasonable option if you value simple, predictable intervals.

  • FSRS. On the open-spaced-repetition benchmark, built from the review logs of thousands of Anki users, FSRS predicted recall more accurately than SM-2.[8] The Anki FAQ says it needs fewer reviews for the same retention.[1] It adapts to your data and lets you choose the retention you want. The numbers are in FSRS vs SM-2.
  • SM-2. The rules fit in a few lines, so you can predict and explain every interval. Most older Anki guides and shared deck settings assume it, and it needs no review history to work well.

Switching to FSRS keeps your full review history, and existing due dates stay as they are unless you choose to reschedule.[3]

Sources

  1. Anki Manual. What spaced repetition algorithm does Anki use? Anki FAQs. Retrieved November 8, 2026. https://docs.ankiweb.net/faqs/what-spaced-repetition-algorithm
  2. Ankitects (2023). Anki 23.10 release notes. GitHub. https://github.com/ankitects/anki/releases/tag/23.10
  3. Anki Manual. Deck options. Retrieved November 8, 2026. https://docs.ankiweb.net/deck-options.html
  4. Ankitects (2026). Anki 26.09.4 source code: default deck options (rslib/src/deckconfig/mod.rs) and the FSRS setting (rslib/src/config/bool.rs). GitHub. https://github.com/ankitects/anki/blob/26.09.4/rslib/src/deckconfig/mod.rs
  5. Woźniak, P. A. (1990). SuperMemo 2: Algorithm. In Optimization of learning (Master's thesis, University of Technology in Poznań). https://super-memory.com/english/ol/sm2.htm
  6. Anki Manual. Studying. Retrieved November 8, 2026. https://docs.ankiweb.net/studying.html
  7. Anki Forums (2023). Does FSRS solve “ease hell”? https://forums.ankiweb.net/t/does-fsrs-solve-ease-hell/38275
  8. Open Spaced Repetition. SRS Benchmark: a benchmark for spaced repetition schedulers/algorithms (results as of November 2024). GitHub. https://github.com/open-spaced-repetition/srs-benchmark/blob/cb44955a3bddce240bb36b481958a67602f15208/README.md