Origin and history of FSRS
FSRS was created in 2022 by Jarrett Ye, a Chinese student researcher passionate about memory algorithms. Dissatisfied with SM-2's limits (Anki's default since 1987), he developed a new model based on real data from hundreds of thousands of Anki users.
The algorithm evolved rapidly: FSRS-3, FSRS-4, then FSRS-4.5 and FSRS-5 (2024). Each version improves prediction accuracy and personalization. In 2024, the Anki team integrated it natively as an alternative to SM-2.
DSR model: Difficulty, Stability, Retrievability
FSRS-5 models memory with three per-card variables, unlike SM-2 which uses just one (the ease factor).
Difficulty represents the intrinsic effort a card demands. A complex-concept card has higher difficulty than a simple-concept one. This parameter is learned from your responses.
Stability measures how long a card's memory will last before you forget. The more successful reviews, the higher stability. A high-stability card can be spaced months out.
Retrievability is the probability you'll remember the card at a given moment. It declines over time along an exponential curve whose shape depends on difficulty and stability.
Why FSRS-5 beats SM-2
SM-2 uses a single ease factor that adjusts after each review. Simple but coarse: a hard card and an old card are treated alike. FSRS-5 separates these dimensions, giving far more accurate predictions.
Benchmarks show that for a 90% retention target, FSRS-5 needs 30 to 50% fewer reviews than SM-2. Concretely, you spend 30 to 50% less time reviewing for the same memorization outcome.
Configurable retention threshold
A FSRS-5 specificity: you can choose your retention target. Default 90%, lower to 80% to review less often (but forget more cards), or raise to 95% for near-perfect retention (at the cost of more reviews).
Diane lets you tune this based on context. For an exam in 2 months, 90% is the right compromise. For one-shot knowledge (a trip), 80% is enough. For long-term mastery (competitive exams), 95% is preferable.
FSRS-5 in Diane vs Anki
Diane uses FSRS-5 natively, no configuration. Anki has used it since 2024 but requires manual activation (Settings > Algorithm > FSRS) and parameter optimization via the Optimize function.
If you use Anki with FSRS-5 enabled and optimized, the algorithm is equivalent to Diane's. If you use Anki with default SM-2, Diane will be more efficient.
Going further
FSRS is fully open source. Parameters, documentation and benchmarks are public on GitHub (open-spaced-repetition/fsrs5). If you're curious about the math, Jarrett Ye has published several articles describing the equations and methodology.