How does FSRS work in Anki? The three numbers behind every review date

How does FSRS work in Anki? It keeps three numbers for every card - difficulty, stability and retrievability - and sets the next due date for the day your chance of recall is predicted to fall to the retention level you asked for. Desired retention is the dial you turn; the parameters behind it are fitted to your own review history.

Everything below is taken from the Anki manual's FSRS section, Anki's FAQ on its scheduling algorithms and the FSRS algorithm wiki as they read on October 23, 2026, plus Anki's release notes. The current stable release is 26.09.3, published on September 23, 2026.

How does FSRS work in Anki? Three numbers per card

FSRS stands for Free Spaced Repetition Scheduler. Anki's FAQ describes the memory state it tracks for each card as three values, which the FSRS project calls the DSR model:

The FAQ puts the split plainly: "The value of R changes daily, while D and S change only after a card has been reviewed." That is the whole mechanism. Stability is a number of days, so a card whose stability is 30 is one Anki expects you to recall with about 90% probability 30 days after the last review - and at the default retention setting, that is roughly when it comes back.

Difficulty is bounded between 1 and 10 in the current algorithm, according to the FSRS wiki, and it is not a measure of how hard the card felt once. It is how resistant the card is to gaining stability at all.

The forgetting curve is what turns stability into a date. In FSRS-6 it is a power function of elapsed time over stability, with its decay constant chosen, in the wiki's words, "to ensure R(S,S) = 90%" - recall sits at 90% exactly when elapsed days equal stability. Change your target away from 90% and Anki solves the same curve for a different point on it.

Desired retention is the one setting most people should touch

The manual is unusually direct about this. Desired retention is "the proportion of cards recalled successfully when they are due", and the guide calls it, in bold, "the most important setting in FSRS. Higher retention leads to shorter intervals and more reviews per day."

The default is 90%, which the manual says "offers a good balance of retention and workload". The cost curve is not symmetric: "Above 90% the workload increases very quickly, and above 97% the workload can be overwhelming," and the manual recommends you "keep it lower than 97%". In our view that asymmetry is the single most useful thing to know about FSRS settings - moving from 90% to 95% buys you five points of recall for a workload increase you cannot predict from the number itself.

Anki will show you your own curve. Help Me Decide, which the manual labels experimental, "allows you to simulate workload at different levels of retention", so you can see your personal retention-workload trade-off in reviews or minutes per day. It replaced the old Compute Minimum Recommended Retention button, which the manual records as "removed as of anki version 25.07" - if a guide tells you to press CMRR, it was written for an older Anki.

Two scoping details matter. Parameters and desired retention are preset-specific, so different decks can run different targets through different presets, and the manual notes that in the latest version of Anki "you can set different values of desired retention for different decks within the same preset."

FSRS parameters come from your review history, not from someone else's settings

The manual's instruction here is a single sentence with no hedging: "Do not change the parameters manually or copy them from someone else."

What the Optimize button does is fit the algorithm to you. The manual says "the FSRS optimizer uses machine learning to learn your memory patterns and find parameters that best fit your review history", drawing by default on the review history of every deck using the current preset. FSRS-6 carries 21 parameters, per the FSRS wiki, and Anki 26.09's release notes record an update of the bundled fsrs-rs library to version 6.6.2, so a current install is scheduling with the FSRS-6 generation of the algorithm.

Optimization is not automatic and does not need to be frequent. "There is no need to optimize your parameters frequently: once every month is sufficient," the manual says, and Optimize All Presets does every preset in one click. If you see "The FSRS parameters currently appear to be optimal", there is nothing to do.

The optimizer also has a diagnostic most people never switch on. Check health when optimizing (slow) reports how well FSRS can adapt to your history, and the manual names the three usual reasons it cannot: a low number of reviews ("less than a few hundred"), misuse of Hard, and not using Again when you forgot. The Hard problem is worth repeating because it quietly poisons every interval: "When you press 'Hard', FSRS assumes you have recalled the information correctly," so pressing Hard on a card you blanked on will push all your intervals too high.

Switching from SM-2 does not touch your backlog unless you ask it to

This is where expectations go wrong. By default, enabling FSRS changes nothing you can see today. Reschedule cards on change is off, and the manual explains that "future reviews will use the new scheduling, but there will be no immediate change to your workload."

Turning it on rewrites due dates, which "will often result in a large number of cards becoming due", and the manual says in bold that "this option is not recommended" when first switching from SM-2. It also warns the option "will add a review entry to each of your cards, and increase the size of your collection", and advises making a backup first if you use it anyway. We would leave it off and let the new scheduling arrive card by card.

Three more constraints are easy to trip over:

• The switch is global. "FSRS can only be enabled globally; you cannot enable it for some presets and disable it for others."
• Every client has to understand it. Anki 23.10, AnkiMobile 23.10 and AnkiWeb support FSRS, and AnkiDroid from 2.17 - "if one of your clients doesn't support it, things will not work correctly."
• Some SM-2 settings disappear rather than being ignored silently: Graduating interval and Easy bonus are among the options the manual says are hidden once FSRS is on. Add-ons that affect intervals should come off too.

Do learning steps still apply with FSRS?

Yes, and they are the one place where your own numbers still override the algorithm. The manual's short guide asks you to "ensure that all your learning and re-learning steps are shorter than 1d and can be completed on the same day", singles out 23h as a bad choice "even though it's less than one day", and offers 10m or 30m as good steps.

It also argues for fewer steps, not more: "Evidence shows that repeating a card multiple times in a single day does not significantly contribute to long-term memory." In the latest version you can hand short-term scheduling to FSRS entirely by leaving the steps field empty, which the manual flags as experimental.

Three things said about FSRS that the documentation does not say

That FSRS is Anki's default. As of the version we checked on October 23, 2026, it is not. The manual still introduces FSRS as "an alternative to Anki's legacy SuperMemo 2 (SM-2) algorithm" that you enable at the bottom of the deck options page, and in Anki's source the FSRS flag is absent from the list of settings that default to true - the rest "default to false". A fresh profile studies on SM-2 until someone flips the switch.

That the newest Anki runs FSRS-4.5. If a guide names an FSRS version older than 6, it is describing an older install. Anki's own FAQ names no version at all, so the release notes are the thing to check - and they put fsrs-rs 6.6.2 in 26.09.

That FSRS cuts reviews by a fixed percentage. Anki's FAQ makes the claim without a number: "With FSRS, users have to do fewer reviews than with Anki's default algorithm to achieve the same retention level." The commonly quoted 20-30% figure is RemNote's own estimate for its own implementation, in its FSRS help article dated August 21, 2026. Both are vendor statements about their own schedulers, and neither is a promise about your deck.

FSRS is not an Anki-only feature

If you keep your source material in a notes app rather than a dedicated flashcard tool, the algorithm has travelled. RemNote, one of the 145 products in our notes and knowledge category as of October 2026, offers FSRS next to the classic Anki SM-2 scheduler. Its help article says "we're currently using FSRS version 6", calls FSRS "a beta feature", and asks for "at least 1,000 reviews with the default weights" before you run its optimizer - a stricter bar than Anki's manual, which only warns that fewer than a few hundred reviews gives the optimizer too little to learn from.

In our view that makes the scheduler a feature to compare rather than a reason to choose: Upfynd lists 318 products in learning and development as of October 2026, and when two of them both say "FSRS", we would ask which version, whether you can optimize on your own review history, and whether desired retention is a setting you control. New ones keep arriving - they show up on our launch boards ranked by upvotes - and if you build one, you can list it on Upfynd for free and say which scheduler it runs.

Turning FSRS on, in order

1. Update every device you study on - the desktop app, AnkiMobile or AnkiDroid 2.17+ - before you touch the setting, since the switch is shared across all of them.
2. Open Deck Options, scroll to the FSRS section at the bottom, and enable it. Leave Reschedule cards on change off.
3. Shorten any learning or relearning step of a day or more to something you can finish the same day.
4. Click Optimize under FSRS parameters, or Optimize All Presets, and enable the health check while you are there.
5. Leave desired retention at 90% for a few weeks, then open Help Me Decide and look at your own workload curve before changing it. Re-optimize about once a month, and resist copying anyone else's parameters.