SmartRecall vs Quizlet: SM-2 Spaced Repetition vs Leitner Boxes (2026)

I'm going to write the comparison I wish I'd read in 2024, when I was a medical student trying to figure out whether the flashcard tool I'd used in undergrad (Quizlet) was the right tool for USMLE Step 1 (it wasn't, but not for the reasons most people think).

I run SmartRecall, so the bias is on the table. I'll try to be fair. Quizlet has roughly 60 million monthly active users. We do not. If you want a single number to anchor scale, that's the one. SmartRecall is not competing for the K-12 vocab-quiz market and we're not going to win that fight even if we wanted to.

What we are competing for is the audience that quietly migrated off Quizlet to Anki over the last five years because Quizlet's algorithm wasn't good enough for exams that require remembering thousands of facts six months from now. That migration is real, it's painful (Anki's authoring wall is its own problem — see my founder essay for that whole story), and the algorithm difference at the heart of it is the actual interesting subject of this post.

The thing nobody tells you about Quizlet's algorithm

Quizlet "Learn" mode is sometimes described as spaced repetition. It is not, in the sense that Anki and SuperMemo and SmartRecall use that term.

Quizlet's Learn mode is an adaptive Leitner-style box system with question-type rotation. Cards you get wrong come back sooner; cards you get right move "up" through stages; the system rotates between question formats (multiple choice, written response, true/false, matching) to vary the difficulty of retrieval. It's a perfectly competent design for a study session, and for a one-week-out vocab quiz it works well.

What it does not do — and this is the load-bearing technical fact for this comparison — is compute a per-card retention probability and schedule the next review at the moment that probability decays to a target threshold. That is what SM-2 does. That is the difference between a study tool and a memory tool.

SmartRecall uses SM-2: every card carries an ease factor, an interval, and a lapse count. After each review, the algorithm updates those three numbers and predicts the optimal next review date — typically days for new cards, weeks for established ones, months for cards you've gotten right repeatedly. This is the same lineage as Anki and SuperMemo, the algorithms that thirty years of cognitive-science evidence have validated for long-horizon retention.

Why does this matter? Because for vocab cramming on a one-week horizon, both approaches give you roughly the same outcome. For a USMLE-style exam where you need to remember 8,000 facts in eight weeks and still have them at exam time, the gap between fixed-stage Leitner and probabilistic SM-2 is not subtle. Cepeda et al. (2008) showed retention improvements of around 200% from properly distributed practice over massed practice. The "properly distributed" part is what SM-2 optimizes for, card by card. Leitner approximates it with fixed boxes.

For a direct comparison of SM-2, FSRS, and Leitner, see SM-2 vs FSRS vs Leitner explained.

The comparison table

What Quizlet wins, plainly

The shared-set library is the real moat. There are millions of community decks on Quizlet — every AP Biology unit, every common-language vocab list, every textbook chapter someone has ever quizzed themselves on. If you're a high schooler studying for tomorrow's Spanish quiz on the imperfect tense, the right answer is almost certainly to search Quizlet, find a deck someone already made, and hit Learn mode. SmartRecall would be overkill and you'd pay for AI generation you don't need.

Quizlet's K-12 quick-study UX is also genuinely best-in-class. Match and Gravity (game modes) are well-designed for short attention spans and have kept a generation of students engaged in repetition they would have otherwise abandoned. The teacher integrations, the class accounts, the way Quizlet slots into a school's existing workflow — none of this is something we're going to rebuild and we shouldn't try.

Annual pricing is also Quizlet's win for casual use. $35.99 a year is roughly $3 a month, less than our Student tier, and for someone who uses flashcards twice a week for general study, that math is correct.

What SmartRecall wins, plainly

For exam prep with a retention horizon of weeks-to-months, SM-2 measurably beats Leitner boxes. This is not a marketing claim, it's the same reason Anki took over medical-school study and Quizlet didn't. If you're going to invest 200+ hours of review time into a deck, the algorithm scheduling those reviews matters, and a probabilistic per-card model will outperform fixed stages every time on the 6-month horizon.

PDF-to-cards solves the authoring wall. Quizlet assumes you have a list of terms and definitions, or that someone else has made one and shared it. For specialty exam prep — pharmacology, pathology, jurisprudence, niche professional certs — the shared decks either don't exist or aren't trustworthy, and typing 8,000 cards by hand is the unpaid labor that breaks most preppers before they get to review even once. Upload a chapter, get cards back in minutes, start reviewing today. That's the workflow change.

Multi-format card generation matters more than it sounds. Cloze deletions test recall in context. Basic Q&A tests bare retrieval. MCQ with realistic distractors tests discrimination (which is what most professional exams actually test). Case analysis tests application. Quizlet's Learn mode rotates question formats on the same flat term/definition pair; SmartRecall generates structurally different cards from the same source, which is closer to how a good teacher would write a problem set.

The honest use-case framing

You have a Spanish vocab quiz on Thursday. Use Quizlet. Find a deck, hit Learn mode, you're done.

You have USMLE Step 1 in eight weeks and roughly 8,000 facts to commit to memory. Use SmartRecall, or use Anki if you have the time and discipline to author cards by hand. Quizlet is the wrong tool for this job and using it will cost you points on test day.

You're learning a foreign language at the conversational level over a year. Either tool works. Quizlet's social features and shared decks probably tip it. If you want the algorithm to actually optimize your reviews over a 12-month horizon though, SmartRecall (or Anki) will give you better recall per minute spent.

You're a teacher assigning vocab. Quizlet, every time. The class infrastructure is built for you.

You're prepping for the Bar, the CFA, 法考, 考研, or any other professional exam where the horizon is months and the volume is thousands of facts. SmartRecall is built specifically for this case. The whole reason the product exists is that I tried to do this on Anki, hit the authoring wall, and rebuilt the workflow around AI generation feeding an SM-2 scheduler.

Why fixed Leitner intervals underperform on long horizons

Quick technical aside, because if you've read this far you probably want it.

Leitner boxes were proposed in 1972 by Sebastian Leitner. The idea: cards live in numbered boxes; correct answers move a card to a higher box with a longer review interval; wrong answers send it back to box 1. The intervals between boxes are fixed at design time — common schedules are 1 day, 3 days, 7 days, 14 days, 30 days.

SM-2, published by Wozniak in 1987, throws out the fixed boxes. Each card has its own ease factor (a multiplier between roughly 1.3 and 2.5) and its own interval, both updated after every review based on how easily you recalled it. A card you find easy might go from a 1-day interval to 6 days to 15 days to 38 days, with the multiplier creeping up. A card you keep failing has its ease factor cut and its interval reset, so it gets hammered until it stabilizes.

The empirical case for SM-2 over Leitner on long horizons is that retention probability is not a step function. It decays smoothly, and the optimal next-review point is different for every card and every learner. Fixed Leitner stages either review too early (wasting time on cards you'd remember) or too late (you've forgotten the card and the next exposure is functionally a new learning event, not a review). On a one-week horizon the loss from this mismatch is small. On a six-month horizon it compounds.

The Cepeda et al. (2008) meta-analysis is the cleanest reference: distributed practice produces retention gains over massed practice that scale with the gap between presentation and test. SM-2 is engineered to land each review inside the gap that maximizes that gain, per card. Leitner approximates it.

What we don't do (yet)

I'll be straight about gaps so you don't get surprised.

We don't have a shared-deck marketplace. I considered it. The decision against was deliberate: shared decks reintroduce the problem the product is built to solve in reverse — you end up drowning in cards you didn't choose, made from sources you didn't read. SmartRecall makes your deck from your source. That's the whole point.

We don't have Quizlet's game modes (Match, Gravity, Live). They're well-built, our users haven't asked for them, and adding them would dilute the product's focus on long-horizon retention.

We don't yet have Android. iOS first, Android in development. Web works on any browser in the meantime.

We don't currently export to Anki. Several power users have asked. It's on the list. If you're an Anki person who wants SmartRecall to do the AI generation step and then hand the deck off to your existing Anki workflow, email me — I want to know how many of you there are before I prioritize it.

CTA

If you're prepping for a high-stakes exam and the authoring wall is what's killing you, start a SmartRecall account — 20 free credits, enough to test the AI generation on one full chapter and decide for yourself whether the cards it produces are the cards you would have written if you'd had the time. If they are, the rest of the math is easy. If they aren't, no harm done and Quizlet (or Anki) is still there.

If you're studying vocab for a quiz next week, just use Quizlet. I mean it. Don't pay us for a job a free Quizlet set does fine.

— Alex

@alexrchen on GitHub · @alexrchen on Product Hunt · [email protected]