If you've ever used Anki, you've experienced SM-2 — the algorithm developed by Piotr Wozniak in 1987. It's effective, but it was designed before we had the computing power or data to do better. Enter FSRS (Free Spaced Repetition Scheduler), an open-source algorithm trained on hundreds of millions of real reviews.
The Forgetting Curve
In 1885, Hermann Ebbinghaus discovered that memory decays exponentially over time — unless you review the material right before you forget it. This is the forgetting curve. Spaced repetition works by scheduling reviews at the optimal moment, reinforcing memory just as it's about to fade.
The challenge is predicting exactly when you'll forget something. SM-2 uses a simple formula based on "ease factors." FSRS uses a machine-learning model trained on real review data to predict your memory state with three dimensions:
- Stability (S) — how long the memory will last after a successful review
- Difficulty (D) — how hard the card is for you personally
- Retrievability (R) — the probability you can recall it right now
FSRS vs SM-2: What's Different?
SM-2 schedules every card using a single "ease factor" that adjusts based on your ratings. This works but it's a crude approximation. FSRS models memory more accurately by tracking stability separately from difficulty.
In independent benchmarks, FSRS achieves ~20% fewer reviews for the same retention rate compared to SM-2. That means less time studying, more time living.
| Feature | SM-2 | FSRS-5 |
|---|---|---|
| Algorithm basis | Heuristic formula | Machine learning model |
| Memory model | 1 dimension (ease) | 3 dimensions (S, D, R) |
| Trained on real data | No | Yes (hundreds of millions of reviews) |
| Review efficiency | Baseline | ~20% fewer reviews |
| Per-user adaptation | Partial | Full (custom weights per deck) |
How Flevro Uses FSRS-5
Flevro runs FSRS-5 natively — not as a plugin or add-on, but as the core scheduler. Every time you rate a card (Again / Hard / Good / Easy), the algorithm updates your personal memory model for that card and schedules the next review.
On our Pro and Premium plans, you can also configure custom FSRS parameters per deck. This means your Japanese vocabulary deck can be tuned differently from your medical terminology deck, because you learn each differently.
What Does "Desired Retention" Mean?
FSRS lets you set a target retention rate — the probability you want to remember a card when it comes up for review. The default is 90%, meaning you'll know the answer ~9 out of 10 times. Set it higher (e.g., 95%) and the algorithm schedules more frequent reviews. Set it lower (e.g., 80%) and intervals get longer — fewer reviews, but more forgetting.
For most learners, 90% is the sweet spot between efficiency and retention. Language learners often drop to 85% for lower-frequency vocabulary; medical students often push to 92–95% for critical recall.
Try It Yourself
The best way to understand FSRS is to use it. Start Flevro for free — no credit card required — and experience the difference after your first month of consistent reviews.