Here is a short list of Anki settings that I think everybody should adapt that can increase your efficiency by an order of magnitude.
- FSRS On. Anki v. 25.07 or higher (FSRS-6).
Both of the above are non-negotiable musts. Google FSRS if you want to learn more about it. It’s good and should be used. Anki 25.07 was the deployment of FSRS-6 which offers significant benefits over previous versions, esp. in the DR<80% region.
- DR (Desired Retention): (Probably) 70%.
The default value of 90% is… horribly inefficient and designed for people trying to pass an upcoming test, not people who want as much vocabulary knowledge in their brain for the least amount of effort possible. The devs have said as much on the Anki forums.
Since Anki’s adoption of FSRS-6, and the versions immediately after succeeding 25.07, the CMRR button was removed in favor of the “Help me Decide” feature. The mathematical formulae in “Help Me Decide” graph is fundamentally broken, making the feature less than worthless as it shows misleading data to the user. However the feature is not going to be fixed because FSRS-6 itself is going to be replaced with a different algorithm which is going to make the entire concept obsolete in the coming year or so. (The graph shows higher efficiency at lower DRs… this is probably true, but is just a coincidence.)
However, you can work around the flawed math by doing the following steps which will then give you the correct output:
(I don't remember which exact version of Anki implemented the "Help me Decide" feature, but I think its was about 3 releases after 25.07. The current release, 25.09.2, has it.)
1) Create some deck. Put cards in it. Do reviews until Anki has data on how well you know that deck.
2) Create a 2nd new empty deck. Use the same Settings Preset as the previous deck.
3) Hit the "Help me Decide" button. Use the following settings:
Days to simulate: 365
Additional new cards to simulate: 99999 (Anything huge is fine)
New cards/day: 20 (Anything >0 is fine)
4) Hit the Simulate button.
5) Make sure you're viewing the "Time / Memorized Ratio" graph.
Wherever is the lowest on the graph, that is the most efficient DR value for you. That is the least time per time to memorize a single unit of information.
It will probably be DR=70%.
You can also, just… try it out. Make a new deck and set it to DR=70 and see how it works and how it compares to your old settings. You'll probably find that you have way less reviews per day for the same number of new cards per day and thus… can simply increase your number of new cards per day to something that gives you faster progress.
After spending a lot of time on the topic, and lots of mathematical analyses of the situation, I believe that for most users, the optimal DR value is probably closer to 60-65%, but 70% is hardcoded as a limit and that probably isn’t going to change (despite the negative effects on the community) until the next scheduling algorithm (which is in active development) is deployed.
So try 70% out. Maybe 75% or 80% are better for you, but probably not.
- Learning Steps: (blank)
It is an undocumented feature in Anki that if you have FSRS turned on and leave this field blank, it will use FSRS to compute the “optimal” learning steps. The reason this is left as a hidden feature is because the FSRS-6 model (and previous versions) simply is inaccurate in the short time-frames of <1 day. So it’s not nearly as good as it is in the multi-day+ region, where it is HIGHLY accurate.
However, despite that, any other setting is even more inaccurate than that, and the flawed FSRS-6 prediction is the best option you have, so use it. You’ll probably just have 1 sub-1-day review anyway.
- Relearning Steps: (blank)
Same as learning steps above, this is a hidden feature for the same reasons and should be enabled for the same reasons.
Using the above settings I was able to maintain somewhere around 230 (!!!) new cards per day over a period of 5 months before eventually succumbing and getting a backlog (currently working on clearing it… going to take a month+… probably going to turn down to ~100 new cards per day in the end). That gave my Anki time around 2hrs/day, so ~50 new cards/day should be ~30min/day which is a reasonable amount of Anki for most students, I suppose. Although that it just a rough estimate and will be different for different people, but it feels manageable and sustainable for long periods of time for most students.
- “Optimize All presets” and calibrating down to DR=70.
Hit this button every now and then. The official dev says “about once a month”. However, Anki has no implementation of actually accounting for calibration inaccuracy of the parameter fitting and also is highly inaccurate when you make sudden drastic changes to DR, so perhaps it is better to only adjust DR by about 1 percent per day, and hit this button every day until you get down to DR=70, and then after that, just hit it about once a week, and then about once a month after that. I could do and show a lot of complicated math about why that’s probably a better idea than just setting DR=70 and doing everything else, but I don’t really feel like spending an hour deriving a bunch about extrapolating error through a high-parameter machine-learning function, but if you “slowly” ease into DR=70 while hitting this button every day or so until it settles in, it will be a much smoother transition than if you were to just… hit this button once a month where you might e.g. have cards be horrendously overestimated lengths and/or optimizing causing hundreds/thousands of cards to appear when you hit the button.
- Burying: all on
This is just common sense.
- New cards/day.
An amount that will cause your time in Anki every day to be about 30min/day at most.
The old common sense was “DR=80-85, 20 new cards per day”. I think the new common sense will become “DR=70, 50 new cards per day”. Of course, it will depend on the person, how much time you want to spend, how maintainable your study routine is, and so on. However, I suspect that somewhere around 50 new cards/day is probably easily doable with the above settings for most everyone. Of course, time will tell if that prediction is true or not.
And I don’t know where else to write it so I’m going to put it here:
Anki, on one hand, has no idea which words you see outside of Anki. But between how long you remember things, and how accurate the FSRS algorithm is, how well the parameter fitting is… if you do reviews outside of Anki, i.e. by reading/consuming Japanese content, seeing the words “in the wild”, you will remember words for longer (from Anki’s POV) and Anki will adapt to this and it will give you longer intervals, allowing you to have more new words per unit time in Anki.
That’s right, if you consume and read a ton outside of Anki, Anki, somewhere in the FSRS-6 parameter fitting and stochastic nature of pass/failing, Anki in some mathematical way “knows” that and gives you longer intervals accordingly. If FSRS-6 thinks you have a 70% chance of recalling a card, then you are almost certain to get 68-70% chance of recalling it, regardless of how often you read/etc. outside of Anki. Although this specifically has not been tested as far as I have seen, just look at this calibration curve and/or look at your actual correct percentage and how close it is to your DR.
What that means is that, if you do a review of a word outside of Anki… Anki kinda knows that, and it counts (in a different and indirect way).
tl;dr:
Anki v. 25.07+
FSRS: on
Learning steps: (blank)
Relearning steps: (blank)
DR: 70% (turn down approx. 1 percent per day until you hit 70)
Optimize All presets: Hit this about once a day until you reach DR=70, and then for about a week after, and then after that, once per month is fine.
Burying: All on.
New Cards/day: A number that ends with you studying about 30min/day inside of Anki, probably somewhere around 50.
Outside of Anki: Read a metric ton.
Edit: Do not do 230 new cards/day. This is not even remotely sustainable. I have since burned out and cannot even do more than 110 reviews/day in Anki anymore. I think some number where your time in Anki each day is <= 30min/day is sustainable. I think that, with the above settings, somewhere around 50 new cards/day is possible and sustainable.
by No-Cheesecake5529