What Desired Retention Should You Choose in FSRS?

Choose the highest desired-retention setting that you can keep up with calmly over ordinary weeks. In FSRS, it is a trade-off: a higher target asks the scheduler to show cards sooner, so you forget fewer cards when due but complete more reviews. It is not a score, a promise, or a number that should be identical for every deck and learner.

Key takeaways

• Anki defines desired retention as the proportion of cards you aim to recall when they are due.
• Anki’s default is 90%; its manual says workload rises very quickly above 90% and can become overwhelming above 97%.
• Use your own sustainable time and review history to compare settings; a simulator is a forecast, not a guarantee.

Scope: this guide explains the choice in current Anki FSRS. It does not promise a particular recall rate, time budget, or matching result in another app. Review history, card quality, honest grading, new-card intake, and missed days all affect the experience.

Start with the meaning, not a magic percentage

Desired retention is the probability you want to recall a card when FSRS schedules it for review. At 90%, the target is a 90% chance at that point; it does not mean you will remember exactly 90% in every session or every exam. Anki calls 90% a good balance between retention and workload, but that is a default starting point, not a universal prescription (Anki Manual: Deck Options, accessed 2026-08-11).

The practical direction is reliable: increasing the setting shortens intervals and increases reviews; decreasing it accepts more forgetting and reduces reviews. The exact curve differs between people and collections. As a target approaches 100%, Anki says workload rises drastically and recommends staying below 97% (Anki Manual: Deck Options, accessed 2026-08-11).

Use 90% as a baseline, then observe

For many people, starting at Anki’s 90% default is sensible because it gives you a baseline with your own cards and grading habit. Keep it long enough to see normal days, not only a backlog day or a week when you changed how many new cards you add. FSRS parameters and desired retention are preset-specific, so clearly different material can be separated instead of forcing one target onto everything (Anki Manual: Deck Options, accessed 2026-08-11).

Before changing the number, check the simpler pressure points. Anki notes that learning new material temporarily increases future reviews; if you consistently learn 20 new cards daily, it gives a rough example of about 200 daily reviews. Reducing new cards can lower burden without changing the recall target (Anki Manual: Deck Options, accessed 2026-08-11).

Make the choice from time you actually have

Use the experimental Help Me Decide view, when available in your Anki version, as a comparison tool. It simulates your personalized retention–workload relationship from your history. Compare a few nearby values and ask two concrete questions: “How many reviews or minutes can I do reliably?” and “Is the added effort worth the extra recall?” (Anki Manual: Deck Options, accessed 2026-08-11).

Treat its output as a model, not a promise. It cannot know a future interruption, whether you will keep grading consistently, whether cards are ambiguous, or whether you will add a new course. A neat graph does not make an unsustainable plan sustainable.

The useful setting is not the one that looks strongest in a simulator. It is the one that still leaves room to review honestly when your schedule is ordinary rather than ideal.

Change one variable at a time

1. Keep a recovery copy. Back up before a major scheduling experiment, especially if you may reschedule due dates.
2. Use a baseline. Start with the default or your current stable value and note a normal week’s workload.
3. Compare nearby targets. Use the simulator, if present, to compare workload and time rather than searching for a perfect percentage.
4. Choose a sustainable value. Prefer a target you can meet through busy periods over a higher target that creates avoidance.
5. Wait before judging. Keep new-card intake and grading habits stable while you observe.
6. Fix cards before blaming FSRS. Repeatedly failed or unclear cards need editing, splitting, or better source learning.

When you change desired retention, Anki normally applies the new scheduling to future reviews without immediately changing due dates. Turning on rescheduling can make many cards due at once, so Anki does not recommend it when first switching from SM-2; it also advises making a backup first (Anki Manual: Deck Options, accessed 2026-08-11).

Keep the evidence feeding FSRS honest

FSRS learns parameters from your review history. Anki advises against copying someone else’s parameters or changing them by hand, because their material and grading history are not yours. It also says low review counts and treating a forgotten answer as Hard instead of Again can make the model fit poorly (Anki Manual: Deck Options, accessed 2026-08-11).

That matters for retention choices. If your record says you recalled an answer when you did not, raising desired retention cannot correct the signal. First use Again for a miss, keep cards specific, and give the optimizer enough honest reviews. See FSRS explained without the maths for the underlying model.

Frequently asked questions

Is 90% the best FSRS desired retention?

No. It is Anki’s default and its documented balance point, not a rule for every learner. Start there if you need a baseline, then compare it with the time you can reliably spend and the importance of the material.

Should I set 95% or 99% for important material?

Importance can justify more review time, but a high setting has a steep workload cost. Anki warns that workload rises quickly above 90% and can be overwhelming above 97%. Test a sustainable increase rather than assuming the highest number is safest.

Does the simulator tell me the correct number?

No. It estimates workload from your own history at different targets. It is useful for comparing options, but it cannot guarantee future recall or account for changes in cards, study time, or grading habits.

Should every deck share one retention target?

Not necessarily. Desired retention is preset-specific. If subjects have very different difficulty or stakes, separate presets can make the workload trade-off easier to understand and adjust.

Your next step

Pick a normal week, not an exceptional one. Keep your current target or Anki’s default as a baseline, use the simulator to compare a small change, and choose the workload you can continue without hiding reviews. If you are changing apps too, first create an Anki backup you have tested.

Primary sources consulted

• Anki Manual: Deck Options and FSRS, accessed 2026-08-11.
• FSRS project: ABC of FSRS, accessed 2026-08-11.
• FSRS project: The Algorithm, accessed 2026-08-11.