FSRS vs Anki's SM-2: what actually changes
If you've used Anki, you've used SM-2 — an interval algorithm published in 1987, elegant for its day, and still the default mental model of 'spaced repetition'. FSRS (Free Spaced Repetition Scheduler) is its open-source successor, and the difference is not a tweak: it is the difference between a fixed formula and a model fitted to you.
What SM-2 does
SM-2 gives every card an 'ease factor' and multiplies the interval by it after each success. Fail a card and the ease drops — permanently, unless you intervene. It has no model of how likely you were to remember; it reacts to grades with fixed arithmetic. Its infamous failure mode, 'ease hell', is a card whose ease has been beaten down until it returns daily forever, teaching you nothing but resentment.
What FSRS does instead
FSRS models each memory with three quantities — difficulty, stability, retrievability — and updates them from every review. Crucially, it is trained: an optimizer fits the scheduler's parameters to your own review log, so the forgetting curve it plans around is yours, not a 1987 average. You choose a target retention (say 85%), and it schedules each card for the moment your predicted recall falls to that line.
In the FSRS authors' published benchmarks across large collections of real review logs, that modeling buys the same retention for meaningfully fewer reviews — the workload difference is the whole point. Anki itself now ships FSRS as an opt-in scheduler, which says what the community concluded.
Why it matters more for Chinese
Chinese multiplies cards: thousands of words, and for each one the separate skills of reading, hearing and producing it. At that scale, scheduling efficiency compounds — a scheduler that saves you a fifth of your reviews saves you weeks per year. It also matters that difficulty varies wildly (你好 and 尴尬 are not the same kind of memory); a model that tracks per-item difficulty handles that spread, where fixed arithmetic punishes you for it.
OptimalChinese runs FSRS-6 natively, schedules each of a word's three faces as its own memory, and can fit the parameters to your accumulated history — the same optimization step Anki users run by hand, built in.
If you're staying in Anki
Turn FSRS on — it is in the deck options — set desired retention near 0.85 to start, and re-optimize monthly once you have a few thousand reviews logged. Your existing history carries over; nothing is lost but the ease factors, which is the good news.
Everything above is built into the app: i+1 sentences, three faces per word, FSRS timing. Try it on your own vocabulary.
Start your free week