FSRS Desired Retention: What Should You Set It To?

If you use FSRS — in Anki or any other app — there's one setting that matters more than all the others: desired retention. It decides how often you review, how much you remember, and whether your daily queue stays sane. The default is 90%, and for most people that's the right answer. But it helps to know why, and when to move it. We ran a year-long simulation with the real FSRS-6 scheduler to put numbers on the trade-off.

What desired retention actually means

Desired retention is the probability that you'll still remember a card at the moment it comes up for review. At 90%, FSRS waits until it predicts a 90% chance of recall, then shows you the card. At 80% it waits longer — until your recall has dropped further — so you review less often and forget a little more.

Two things follow from that:

• It isn't your overall score. Between reviews your recall is higher than the target — it starts near 100% right after a review and only sinks to the target at the next one. So at 90% desired retention, the share of your deck you could recall on any given day is noticeably above 90%.
• It's a workload dial. A higher target means shorter intervals and more reviews. The relationship isn't linear: each extra point costs more than the last.

The Anki manual puts the warning plainly: above 90% the workload increases very quickly, and above 97% it can be overwhelming.

What happens at each setting: a year-long simulation

To make the trade-off concrete, we simulated a student who adds 20 new cards every day for a year — 7,300 cards — using the FSRS-6 scheduler with its default parameters. Each review succeeds with exactly the probability the model predicts, so the only thing that changes between runs is the desired retention. Results are averaged over five runs.

Read it from the default outwards:

• 90% → 95% adds about a third to your daily reviews and 45% more total work, to remember about 2.5% more of your cards.
• 90% → 97% roughly doubles your total reviews for about 3.5% more cards remembered.
• 90% → 85% saves about 15% of your daily reviews and costs about 2.5% of your cards.

That's the whole argument for the default: past 90%, you pay a lot of extra review time for very little extra memory. Below it, you save time but the losses start to add up.

How we ran it: FSRS-6 via the open-source ts-fsrs library, default parameters, no interval fuzz, 20 new cards a day for 365 days, success probability equal to predicted recall, "Good" on success and "Again" on failure, same-day relearning resolved immediately. It's a model of an idealised learner — your numbers will differ — but the shape of the trade-off is what matters.

What should you set it to?

• Most people, most of the time: 90%. It's the default for good reason — the best balance of workload and memory in the table above.
• Exam in the next few weeks, material you can't afford to miss: 93–95%. Worth the extra reviews for a short, high-stakes window. Keep it to the decks that matter, not everything.
• Big backlog, busy season, or low-stakes breadth material: 80–85%. You'll forget a bit more but keep the habit alive — and a queue you actually clear beats a perfect one you abandon.
• Almost never: above 97%. The workload climbs steeply and most apps (including Anki's guidance) advise against it.

Two practical tips. Change retention per deck, not globally — your core exam subject and your casual vocabulary deck don't need the same target. And give a new setting a couple of weeks before judging it; the queue takes time to settle.

Common mistakes

• Setting 99% "to be safe". The table shows why this backfires: 97% already doubles the work of 90% for a sliver more memory, and the curve only gets steeper above it — a queue that eventually breaks your habit.
• Lowering retention to fix a backlog, then forgetting to raise it. Fine as a temporary measure; put it back once you're caught up.
• Confusing retention with your success rate on hard cards. Your "true retention" statistic will hover around the target across your whole collection; individual difficult cards will fail more often, and FSRS already accounts for that.
• Tweaking weekly. Retention is a strategy setting. Pick one, then spend your energy on reviewing.

Setting it in Anki

In Anki: open a deck's Options, enable FSRS in the FSRS section, and set Desired retention (0.90 by default). Anki also offers tools to simulate workload at different retention levels. Once you have a few hundred reviews, use Optimize to fit FSRS's parameters to your own memory — that personalises the intervals without changing your retention target. (New to FSRS? Start with FSRS vs SM-2.)

Setting it in StudyTab

StudyTab schedules with FSRS-6 and gives you four presets — Relaxed (80%), Balanced (90%), Intense (95%) and Exam Prep (97%) — or any value from 70% to 97% on the slider. Study modes set retention per deck (for example 85% for a relaxed deck and 92% for an exam deck), and the schedule calendar shows the reviews due over the next 30 days so you can see a change before it bites. (How many new cards that load supports: how many flashcards per day.)

Bottom line

Keep desired retention at 90% unless you have a specific reason not to. Nudge it up to 93–95% for a short, high-stakes run before an exam, down to 80–85% when you need to protect the habit — and treat anything above 97% as a trap. In our simulation, the jump from 90% to 97% doubled the work for 3.5% more memory.

Want the scheduling handled for you? Build a deck and let FSRS pace it.