FSRS-6 vs SM-2: why the scheduler upgrade matters
Rahul Ragi
18.4.2026
FSRS-6 vs SM-2: why the scheduler upgrade matters
If you opened Anki on a new machine in 2026 and did not touch the deck options, you might still be running SM-2. The algorithm has been Anki's default for two decades, it still ships in a lot of third-party tools, and most community decks assume it. FSRS-6 is the upgrade — trained on hundreds of millions of real reviews, shipped as the current Anki default for new profiles, and available as an opt-in for everyone else. The practical question is whether switching is worth the one-time friction.
Short answer: yes for anyone studying seriously past a four-week horizon; not really for casual study on a single subject. The long answer is worth understanding, because most comparison articles overstate the delta by citing benchmarks at a retention target that is not necessarily the one you should be using.
What is FSRS-6 in one paragraph?
FSRS stands for Free Spaced Repetition Scheduler. Version 6 shipped in late 2025 and is trained on roughly 700 million reviews contributed by about 20,000 volunteer Anki users (fsrs4anki wiki, checked 2026-04-18). It models three things per card: stability (how well the memory holds over time), difficulty (how hard the card is for you specifically), and retrievability (your current probability of remembering it). Every review updates the model; the next-due interval is whatever keeps retrievability at your chosen retention target, typically 90%.
What is SM-2 and why is it still the default anywhere?
SM-2 is the SuperMemo-2 algorithm, published by Piotr Wozniak in 1987. It uses a single "ease factor" per card plus a small set of hand-picked multipliers: rate a card "Good," next interval is previous interval × ease factor. Rate it "Hard" and ease factor drops a little. Rate it "Again" and ease factor drops a lot and interval resets.
SM-2 is still the default in places for two reasons. First, inertia — two decades of tooling, deck exports, and scheduling assumptions are built around it. Second, simplicity — the algorithm has maybe six parameters total, which makes it trivial to implement and intuitive to debug. FSRS-6 by contrast has seventeen weights that a trained optimizer tunes per user.
SM-2 is adequate for casual study. If you review 50 cards a day on one subject for a six-week semester, SM-2 will get you to the exam. The failure mode is not catastrophic — it is gradual. Intervals grow too fast for cards you find hard and too slow for cards you find easy, and the cumulative cost is more reviews than you strictly needed.
Where does FSRS-6 actually outperform SM-2?
The headline benchmark everyone cites: FSRS reduces the number of reviews required at matched retention by 20–30% (fsrs4anki wiki, checked 2026-04-18). That number is real but it is also load-bearing on an assumption most articles skip: the benchmark uses a 90% retention target.
Three places the delta is largest in practice:
- Long horizons. SM-2's ease factor has no memory of why a card was hard — it just notices you struggled. FSRS-6 models stability as a function of prior reviews, so a card you kept failing will be scheduled more conservatively than a card you just happened to miss once on a bad day. The difference compounds over months. For anyone studying toward a one-year horizon (USMLE Step 1, bar exam, comprehensive finals), FSRS's 20–30% review reduction is conservative.
- Large decks. On a 500-card deck the absolute number of saved reviews is small. On a 20,000-card deck — AnKing, a full language vocabulary, a multi-year knowledge base — those percentages translate into hours per week. Scale is where the algorithm upgrade becomes visible.
- Mixed-difficulty decks. SM-2 applies the same ease-factor mechanics to every card. FSRS-6 models difficulty separately and schedules the hard cards more often without punishing the easy ones. If your deck has a bimodal distribution — half the cards you nail, half you struggle with — FSRS-6's gain is larger than the average benchmark suggests.
Where does the benchmark overstate?
Here is the contrarian read: the 20–30% figure assumes you are optimising for 90% retention. Push your retention target higher — say 95% for an exam you really cannot miss — and the gap narrows because any scheduler at that target is reviewing you close to the floor of the forgetting curve. Push it lower — 80%, which is defensible for casual self-study — and the gap narrows again because SM-2's imprecision matters less when you are already accepting more forgetting.
The sweet spot for FSRS's advantage is 88% to 92% retention, which happens to be where the algorithm was tuned. If you have consciously chosen a different target, the improvement you will actually see is different from the advertised benchmark.
Most learners have not consciously chosen a target; they are running whatever the default is. At Anki's 90% default, FSRS-6 is the better algorithm. That framing is accurate in the common case — but you can read "20–30% fewer reviews" into an internal forum post and arrive at an expectation that does not match your own retention preference. Know the target you are actually optimising for before you get excited about the delta.
When is SM-2 good enough?
Named concession: SM-2 is adequate for short-horizon, single-subject, low-stakes study. If you are learning 200 Spanish vocabulary words for a two-month trip, SM-2 gets you there. If you are drilling 80 terms for a Friday quiz, SM-2 gets you there. The algorithm becomes hostile only at scale — when your deck is large, your horizon is long, or your workload is mixed across subjects with different difficulty profiles.
The failure mode of staying on SM-2 in the hostile regime is not a disaster. It is a slow tax. You will review more than you needed to, you will forget cards you would have kept if the scheduler had spaced them differently, and you will not know what you lost because the counterfactual is invisible. The argument for upgrading is not that SM-2 is broken; it is that there is a measurable better option and the one-time switching cost is low.
Does Anki default to FSRS or SM-2 now?
As of Anki 23.12 and forward, FSRS-6 is available as the scheduler and is the default for new profiles on recent installs (Anki manual — deck options, checked 2026-04-18). Existing profiles continue on whatever they had unless you flip the switch in deck options. If you installed Anki three years ago and have never been to deck options since, you are almost certainly still on SM-2.
NeverCram ships FSRS-6 as the default for every user with no opt-in required. This is intentional — most learners will not change a default, so whichever scheduler is on at signup is the one they actually run.
How do you switch, and what breaks?
In Anki: Deck options → Advanced → enable FSRS, then optionally run the FSRS optimiser on your review history to tune the seventeen weights to your specific retention pattern. The optimiser takes a few seconds to a minute depending on how large your review log is. Existing scheduling state is translated; intervals will recalibrate over the first week of reviews but no cards are lost.
What breaks: nothing in the deck itself. What feels different: due counts may spike or drop in the first few days as FSRS-6 rescores the backlog. This is the algorithm correcting SM-2's cumulative drift, not a bug. Give it a week before judging.
What this means for your study choices
If you are already on FSRS-6 (new Anki install, NeverCram, or any modern scheduler-first tool), you do not need to do anything. If you are on SM-2 and studying at scale or toward a long horizon, flip the switch this week — the one-time friction is small and the cumulative retention advantage compounds. If you are on SM-2 and studying casually, it is a preference call, not a requirement.
The deeper shift is that "FSRS vs SM-2" stopped being a live debate once FSRS-6 became the default in the tools that matter. The real comparisons in 2026 are between tools that both run FSRS-6 — authoring workflows, session design, platform support. Picking your scheduler is table stakes. Picking your tool is the interesting decision.
Last updated: 2026-04-18. Drafted with AI assistance, edited and fact-checked by Rahul Ragi. Sources verified: fsrs4anki wiki, Anki manual.