Guide

FSRS, the modern spaced-repetition algorithm

FSRS is the open-source, machine-learning scheduler now built into Anki. Here is what it models, how it beats SM-2, and what it means for your reviews.

Part of the Spaced repetition: the complete guide guide.

For most of spaced repetition’s history, the scheduling algorithm barely changed. SM-2 from the 1980s was the default everywhere. FSRS is what finally replaced it, and it is now built into Anki and a growing list of other tools. This guide explains what it does differently and why it matters.

A flat editorial illustration of a soft blue curve being predicted, with a faint dotted continuation and small confidence marks, on a pale desk surface

What FSRS is

FSRS stands for Free Spaced Repetition Scheduler. It is open-source, community-built, and unlike SM-2 it is a genuine machine-learning model fitted to how people actually remember. As of Anki 23.10 it ships in Anki itself (and AnkiMobile, AnkiWeb, and AnkiDroid), so switching to it is a settings toggle rather than an add-on.

The idea: model memory directly

SM-2 schedules by rules of thumb (multiply the interval by an ease factor). FSRS instead models the memory itself, using three quantities for every card:

  • Retrievability — the probability you could recall the card right now. This falls over time, following the forgetting curve.
  • Stability — how slowly that probability decays. A stable memory loses retrievability slowly; a fragile one drops fast.
  • Difficulty — how hard this particular card is for you, which affects how much each review boosts stability.

Every time you review, FSRS updates these numbers from your answer and from data across millions of reviews. Then it does the thing SM-2 never could: it schedules the next review to land when your retrievability is predicted to hit a target you choose.

Why that is better

The practical payoff is that you tell FSRS your desired retention (say, 90%), and it schedules each card to keep you near that level. Because it predicts forgetting per card rather than applying a blanket multiplier, it tends to deliver the same retention with fewer reviews, or higher retention for the same number, compared with SM-2. Less wasted effort on cards you already know, less leakage on cards you are about to forget.

It also avoids SM-2’s well-known failure modes, like the “ease hell” spiral, because there is no ease factor to drift. The schedule comes from the memory model, not from accumulated button presses.

The honest caveats

FSRS is the current state of the art, but it is not magic:

  • It is still card-level scheduling. It optimizes when to show each flashcard; it does nothing for material that is not in cards.
  • It needs a bit of review history to personalize. Early on it uses sensible defaults and improves as it learns your patterns.
  • Self-graded recall is still the input, so the testing effect and honest grading still matter as much as the algorithm.

What it means for you

If you drill flashcards in Anki, turning on FSRS is one of the highest-value, lowest-effort upgrades available: same habit, fewer reviews, better retention. If you are choosing a tool, strong FSRS support is a point in a tool’s favour.

But the algorithm only matters once you are committed to flashcards. If your studying is books, notes, and courses rather than cards, no card scheduler touches it. That gap, scheduling whole sources rather than cards, is what Memset is for. The complete guide lays out where card-level algorithms like FSRS fit and where a planner does instead.