If you use Anki for language learning, you review flashcards every day. Those reviews are scheduled by an algorithm — a mathematical model that predicts when you’ll forget each card and schedules a review just before that moment. The better the algorithm, the fewer reviews you need for the same retention. Fewer reviews means less time spent reviewing and more time available for new learning.

For over 30 years, that algorithm was SM-2. Since 2022, there’s a significantly better option: FSRS. If you haven’t switched yet, you’re spending 20-30% more time on reviews than you need to.

Here’s the science behind both algorithms and why the upgrade matters.

SM-2: The Algorithm That Started Everything

In 1987, a Polish computer science student named Piotr Wozniak published the SuperMemo 2 (SM-2) algorithm — arguably the most influential piece of software in the history of self-directed learning. SM-2 was the first practical implementation of spaced repetition: the principle, discovered by Ebbinghaus in the 1880s, that reviewing material at expanding intervals dramatically improves long-term retention compared to massed practice.

SM-2’s model is elegantly simple. For each card, it tracks:

When you review a card, you rate it on a scale (Again, Hard, Good, Easy in Anki’s implementation). Based on your response: - If you fail, the card resets to a short interval - If you succeed, the next interval = previous interval x EF - The EF adjusts slightly based on your performance

The result: cards you find easy get reviewed less frequently (intervals grow fast), cards you find hard get reviewed more (intervals stay short or shrink). Over time, each card settles into a review schedule roughly matched to your forgetting rate for that item.

Why SM-2 was revolutionary: Before SM-2, students either reviewed everything at fixed intervals (inefficient) or relied on intuition about what they’d forgotten (unreliable). SM-2 automated the scheduling decision, making it possible for anyone to manage thousands of cards without manually deciding when to review each one. Anki adopted SM-2 as its default algorithm, and through Anki, SM-2 became the scheduling engine behind millions of language learners’ daily practice.

The Limitations of SM-2

SM-2 has worked well for decades, but it has fundamental limitations that become apparent with modern understanding of memory:

1. One-size-fits-all model

SM-2 uses the same mathematical model for every learner. The Easiness Factor adjusts per card, but the underlying assumptions about how memory decays don’t vary between individuals. In reality, people differ significantly in their baseline forgetting rates — a card that one learner needs to see every 10 days might be stable for 20 days in another learner’s memory. SM-2 has no mechanism to learn these individual differences.

2. No explicit forgetting curve model

SM-2 doesn’t model the actual probability that you’ve forgotten a card at any given moment. It schedules reviews based on heuristic rules (multiply the interval by EF), not based on a mathematical model of memory decay. This means it can’t tell you “there’s an 85% chance you still remember this card” — it can only say “the interval says to review today.”

3. Binary success/failure

When you forget a card in SM-2, the interval resets to near zero regardless of how long you’d been reviewing it. A card with a 6-month interval that you fail gets treated almost the same as a brand-new card. This is overly punitive — failing a well-established card after a long interval is qualitatively different from failing a new card, and the optimal response should differ accordingly.

4. No retention targeting

SM-2 doesn’t let you specify your desired retention rate. Some learners want 95% retention (accepting more reviews); others would accept 85% retention for dramatically fewer reviews. SM-2 doesn’t have a mechanism for this trade-off — the retention rate is an emergent property of the algorithm’s parameters, not a configurable target.

FSRS: The Modern Alternative

FSRS (Free Spaced Repetition Scheduler) was developed by Jarrett Ye, published at the ACM SIGKDD Conference in 2024, and integrated into Anki starting with version 23.10. It represents a fundamental rethinking of how spaced repetition scheduling should work.

FSRS is built on the Three-Component Model of Memory, which tracks three variables for each card:

1. Stability (S)

Stability represents how deeply a memory is encoded — specifically, the time it takes for the probability of recall to drop from 100% to 90%. A card with S = 30 days means that 30 days after your last review, you have a 90% chance of recalling it. High stability = slow forgetting = long intervals.

2. Difficulty (D)

Difficulty captures the intrinsic difficulty of a card for you specifically. Unlike SM-2’s Easiness Factor, FSRS’s Difficulty parameter is informed by the full history of your interactions with the card and by machine learning models trained on your review patterns.

3. Retrievability (R)

Retrievability is the estimated probability that you can recall the card right now. It decreases over time according to a forgetting curve shaped by Stability. This is the variable that SM-2 completely lacks — FSRS can tell you, at any moment, the estimated probability that you still remember each card in your collection.

How FSRS schedules reviews

When you review a card, FSRS updates S and D based on your response. It then calculates the optimal next review date to maintain your target retention rate — a number you configure (default 90%). If you want 90% retention, FSRS schedules the review so that R will be approximately 90% when the review comes due. If you set 85% retention, intervals stretch longer; at 95%, intervals compress.

This is fundamentally different from SM-2. Instead of arbitrary multipliers, FSRS solves a precise mathematical question: “Given this card’s stability and my target retention, when will the probability of recall drop to exactly that target?”

The personalization advantage

FSRS uses machine learning to learn your memory patterns from your review history. After you’ve completed a sufficient number of reviews (~1,000), FSRS can optimize its 19 parameters to match your individual forgetting curves. This personalization is something SM-2 cannot do at all — it uses fixed formulas that don’t adapt to individual differences in memory.

The Numbers: How Much Better Is FSRS?

The evidence is compelling:

Ye (2024) reported that FSRS achieves the same retention rate as SM-2 with approximately 20-30% fewer reviews. For a learner doing 100 reviews per day under SM-2, that’s 20-30 reviews eliminated daily — roughly 10-15 minutes saved, every day, indefinitely.

The practical impact over time: At 15 minutes saved per day, that’s 91 hours per year — time that could be spent on 91 hours of comprehensible input, conversation practice, or any other acquisitional activity. For language learners who already struggle with review volume (the “Anki burnout” problem), this reduction is the difference between sustaining the habit and abandoning it.

Benchmark studies on the open-spaced-repetition project (GitHub) have compared FSRS against SM-2 and several other scheduling algorithms across large datasets of real user reviews. FSRS consistently achieves better log-loss scores (a measure of prediction accuracy) — meaning its predictions of whether you’ll remember a card are more accurate than SM-2’s scheduling heuristics.

The contrasting perspective: Some long-time Anki users argue that SM-2 is “good enough” and that the optimization gains from FSRS are marginal for learners with small collections (<2,000 cards). There is some truth here — the benefits of FSRS scale with collection size and review volume. A learner with 500 cards doing 30 reviews per day will see modest savings. A serious language learner with 10,000+ cards doing 150+ reviews per day will see dramatic improvements. Additionally, SM-2 has decades of user experience and community knowledge behind it, making it the “safer” choice for beginners who prefer simplicity.

How to Switch to FSRS in Anki

The migration is straightforward and non-destructive:

Step 1: Update Anki to version 23.10 or later (check ankiweb.net for the latest version).

Step 2: Go to Deck Options (gear icon on your deck, or Tools > Preferences on desktop).

Step 3: In the deck options, find the FSRS section. Toggle “Enable FSRS” on.

Step 4: Click “Optimize” to train FSRS on your existing review history. This analyzes your past reviews to personalize the algorithm’s parameters. If you have fewer than 1,000 reviews, FSRS will use sensible defaults and optimize once you have more data.

Step 5: Set your desired retention rate. The default is 0.90 (90%). This means FSRS will schedule reviews so that you have approximately a 90% chance of recalling each card when it comes due. Adjust based on your priorities: - 0.85 (85%): Fewer reviews, more forgetting. Good for large vocabularies where some forgetting is acceptable. - 0.90 (90%): The sweet spot for most language learners. Recommended default. - 0.95 (95%): Many more reviews, minimal forgetting. Only justified for high-stakes material.

Step 6: Review normally. FSRS will immediately begin scheduling using its model. Your existing cards’ histories are preserved — FSRS uses them to estimate each card’s current Stability and Retrievability.

Important: After switching, you may notice that some cards’ intervals change significantly. Cards that SM-2 was scheduling too aggressively (intervals too long) will get shorter intervals; cards that SM-2 was scheduling too conservatively (intervals too short) will get longer intervals. This is FSRS correcting SM-2’s inaccuracies — the adjustment period typically lasts 2-4 weeks.

Configuring Retention: The Trade-Off

The ability to set a target retention rate is FSRS’s most powerful feature for language learners, because it makes an implicit trade-off explicit:

Target Retention Reviews/day (approximate) Forgetting rate Best for
85% Lowest ~15% of cards forgotten per cycle Large passive vocabulary, recognition-only cards
90% Moderate ~10% forgotten per cycle Most language learners (recommended)
95% Highest ~5% forgotten per cycle Small decks, production cards, high-stakes material

The 90% default means that roughly 1 in 10 cards will be “forgotten” when they come up for review. This is not a failure — it’s optimal. Research on the desirable difficulties framework (Bjork & Bjork, 2011) shows that some retrieval difficulty actually strengthens long-term retention. If you never forget anything in your reviews, your intervals are too short and you’re over-reviewing.

The Bottom Line

SM-2 was a breakthrough in 1987. FSRS is a breakthrough in 2024. Both implement the same fundamental principle — review at the moment of near-forgetting — but FSRS does it with a more accurate model of memory, personalization to your individual learning patterns, and the ability to target a specific retention rate.

If you use Anki for language learning, switching to FSRS is one of the highest-impact, lowest-effort improvements you can make. It takes five minutes to enable, requires no changes to your cards or study habits, and saves you 20-30% of your review time — every day, for as long as you use the tool.

The 15 minutes you save daily is 15 minutes you could spend shadowing, reading, or having a conversation in your target language. Over months and years, that compounds into something significant.

The algorithm is the engine. Upgrade the engine.

This article is part of the series “The Science of Language Learning” — where we break down what research actually says about how adults acquire languages, and how to use that science to learn faster.

Previous in the series: Anki: The Complete Setup Guide for Language Learners in 2026

Next in the series: The Best Language Learning Apps in 2026: An Evidence-Based Review

References: