Anki's actual default settings, from the source

I went looking for a list of Anki's default settings and couldn't find one at all, right or wrong.

The manual describes what every option does, in careful prose, and mostly declines to say what it's set to. So people guess, and then other people quote the guesses. This page is the values, each one traced to the line of code that sets it.

This table documents Anki 26.08.1

Anki 26.08.1. That's the string in .version at the root of the repository, on commit 20c475f. It bundles fsrs-rs 6.6.2, pinned in Cargo.toml. That's the code for FSRS, an open-source algorithm that predicts when you're about to forget each card.

A defaults page without a version is worthless in about a year, and most of the ones I found don't carry one. These values move: the FSRS parameter count changed in 25.06, and Compute Minimum Recommended Retention was pulled in 25.07.

To check Anki's factory values yourself without reading Rust: make a new profile (File, Switch Profile, Add) and open any deck's options. A fresh profile is the factory state.

Is FSRS on by default in Anki?

No, and the FSRS default is the one people get wrong, and they get it wrong confidently. Plenty of pages say FSRS "became Anki's default scheduler in version 23.10". A new Anki collection schedules with SM-2, the older formula-based scheduler, and has done since 23.10 shipped.

The manual is unambiguous about the relationship. FSRS is "an alternative to Anki's legacy SuperMemo 2 (SM-2) algorithm". The setup instructions tell you to enable it yourself, in the FSRS section where the deck options page ends.

There's a whole subsection titled "Before Enabling". You don't write a Before Enabling section for something that arrives enabled.

Anki's source code settles whether FSRS ships on. Anki keeps its on/off collection settings in rslib/src/config/bool.rs, where one match statement decides what an unset key returns:

BoolKey::Fsrs is declared 30 lines above and is absent from the true list. It falls through to other, and gets false.

Its neighbour LoadBalancerEnabled is in the true arm. So the load balancer is spreading your due dates by default, even while the scheduler it was built alongside sits switched off.

The deck options screen agrees with the Rust. The FSRS switch itself is drawn as <SwitchRow bind:value={$fsrs} defaultValue={false}>, so Anki's own revert arrow considers off to be the factory position.

FsrsHealthCheck is in the true arm too, and that one is a trap. The deck options screen draws its toggle with defaultValue={false}, so the collection key and the switch disagree.

The "FSRS is off" correction gets overstated in the other direction too. So: FSRS is bundled, free, first-party, maintained in-tree, and about four clicks away.

Anki's position is that you should choose it, and 23.10 is genuinely when it stopped being an add-on. The accurate phrasing is "built in since 23.10". If you want the argument for flipping it, I made it in how FSRS schedules your reviews.

Every default, traced to the line of code that sets it

The deck options screen in screen order, then four rows that live elsewhere. The manual column says whether the Anki manual states that value anywhere, which is the column that surprised me.

One row is deliberately thin. FSRS parameters ship empty and get filled by the optimizer, and the 21 values it starts from belong in their own reference rather than a cell in this table.

Four rows sit outside the deck options screen, so the screen itself renders 46 of these. The load balancer, the 4am rollover and the learn-ahead limit are collection-wide. And new_per_day_minimum is a struct field with no control at all.

For the 46 on the screen, my test was whether Anki draws a revert arrow next to the row. The arrow is the program telling you it knows the default.

That test has one blind spot I had to patch by hand. The parameter search box gets no arrow, because its fallback is computed from your preset name instead of stored. Empty means preset:"<your preset>" -is:suspended, so by default the optimizer trains on that preset's cards and skips suspended ones.

One caution if you go source-diving. The deck config is defined twice, in a protobuf struct and an older schema11 one, and they agree on every value I checked. (Protobuf is the format Anki uses to define its data.)

But protobuf zero-values aren't Anki's defaults: LEECH_ACTION_SUSPEND = 0 sits at the top of that enum and isn't what a new deck gets. The Rust is the authority.

The manual gives no value for 29 of the 50 defaults

Count the No column and you get 29 of 50. The manual states a value for 19. The remaining 2 are the FSRS switch, which the manual implies without stating, and the leech action, where the manual states the wrong one.

Of the 19 it does state, 6 are looser than a number. The review sort order is called "the default order" without naming it. The maximum interval is given as "100 years" rather than 36,500 days. And four switches are described without saying which way they ship.

The pattern isn't random. Everything under Advanced, the SM-2 multipliers, is documented with its number. (Ease is the multiplier that stretches each interval.) They are starting ease 2.50, easy bonus 1.30, hard interval 1.20, interval modifier 1.00, new interval 0.00. Everything a beginner touches first is documented without one. The Learning Steps section runs 400 words, uses 1m 10m 1d as its worked example, and never says the real default is 1m 10m.

The documented-versus-undocumented split is backwards for how people actually get hurt. The Advanced numbers are the ones the manual tells you to leave alone, and it's right. Changing starting ease or the interval modifier by hand is the classic way to break a deck, which is most of why FSRS exists. The undocumented ones are the four that decide whether you're still using Anki in March.

• New cards/day, 20. The one to change, and downward. See the arithmetic below.
• Learning steps, 1m 10m. Fine as shipped. Worth noting the manual's worked example uses 1m 10m 1d, and a day-long step is the one thing it later tells FSRS users to avoid.
• Leech action, Tag Only. Worth knowing about before your deck has 200 of them, because nothing visibly happens when a card becomes a leech (a card you've failed over and over).
• Bury siblings, all three off. If you use reverse cards or cloze (a sentence with the word blanked out), you'll see both halves of a note in the same session by default. That makes the second one much easier than it should be.

Then there's a layer below that, constants no setting exposes. Since they're what actually moves your intervals, they belong in a reference page:

• Ease adjustments per answer. Again subtracts 0.20, Hard subtracts 0.15, Easy adds 0.15. Good leaves ease alone.
• Minimum ease 1.30. Ease cannot fall below it. From 2.50 that's exactly 6 Agains, or 8 Hards, and then a card sits on the floor.
• Interval fuzz. This is a small random spread, so cards don't all land on the same day. Nothing below 2.5 days. At or above it, the spread starts at 1 day. Add 0.15 for every day between 2.5 and 7, 0.10 for every day from 7 to 20, and 0.05 for every day past 20. The bands stack instead of replacing each other, so a 10-day interval lands within about 2 days either way and a 1-year interval within about 20.
• The leech counter repeats. It fires at the threshold, then every half-threshold after, rounded up. On the default 8 that's lapses 8, 12, 16, 20.

And one place where the manual states the wrong value outright. Its Leeches page says that when the lapse counter "reaches 8, Anki tags the note as a leech and suspends the card." The shipped default doesn't suspend.

leech_action is LeechAction::TagOnly, declared with #[default] on the schema11 enum. The answering code suspends only inside if next.leeched() && self.config.inner.leech_action() == LeechAction::Suspend.

Tag only means the card keeps coming back, which is why people report decks full of leech-tagged cards they never noticed. The first step in dealing with one is knowing Anki didn't quietly deal with it for you.

The default new-card limit fills the default review cap

The manual gives one workload example, and it's a good one: "If you are consistently learning 20 new cards a day, you can expect your daily reviews to be roughly about 200 cards/day." That ratio, 10 reviews per new card, is the honest planning number for Anki. It gets mangled into a "10x rule" attached to things it was never about.

Put it next to the shipped values. New cards/day defaults to 20. Maximum reviews/day defaults to 200.

Anki 26.08.1, factory settings

0

new cards/day, the default

0

reviews the manual says that settles at

0

maximum reviews/day, the default

The defaults are internally consistent, which is clearly deliberate. The consequence is that a learner who accepts both and studies daily arrives at the review cap and stays pinned to it. Every day above 200 due cards silently defers the excess, and Anki mentions it only on the congratulations screen.

The number that matters for a beginner hides in the manual's 10-to-1 ratio. 20 new cards a day is a commitment to roughly 200 reviews a day, for as long as you keep adding. At 10 seconds a card, that's over half an hour daily, and 10 seconds is optimistic for a language card with audio.

I think it's a reasonable ceiling and a bad starting point, and it's the first setting I'd change.

Two smaller results. At starting ease 2.50 with a 1-day graduating interval and no fuzz, a card answered Good every time needs 12 reviews to pass the 36,500-day maximum. Collecting them would take 108.8 years, against a cap of 99.9.

The maximum interval is a ceiling on what the multipliers can propose, and studying will never walk a card up to it. The easy bonus compounds too: 5 straight Easy answers land a card 1.30⁵, about 3.7 times, further out than 5 straight Goods.

Where Anki's defaults touch what I build: Verbamor has no deck options screen, and schedules cards without asking you to pick a starting ease. That costs something real.

Anki's 46 deck-option knobs are why it can be tuned to any subject on earth, and why a beginner's first week goes on the manual. I chose fewer because the app only has to schedule vocabulary from your own lessons. If you want the knobs, keep Anki: free on desktop, US$24.99 once on iOS, and better at being Anki than anything else is. I went through that decision in is Anki worth it.

---

Sources