BharatNotes Yes — FSRS-6 (default in Anki 25.07+, shipped July 2025) needs roughly 20-30% fewer reviews than SM-2 for the same retention target, based on ~700M-review training data. For a UPSC aspirant with 3,000-5,000 cards over 12 months, that is ~90 saved hours. Switch, set desired retention to 90%, and re-optimise weights after 1,000 reviews.
What FSRS actually is
FSRS — Free Spaced Repetition Scheduler — is a machine-learning-based scheduling algorithm developed by the open-spaced-repetition project on GitHub. It replaced Anki's three-decade-old SM-2 algorithm as the default scheduler in Anki 25.07 (released July 2025), having first shipped as an opt-in in Anki 23.12 (December 2023).
Unlike SM-2, which models each card with a single 'ease factor', FSRS models three separate variables for every card:
| FSRS variable | What it captures |
|---|---|
| Difficulty | How inherently hard this specific card is for you |
| Stability | How long the memory will last before retrievability drops to 90% |
| Retrievability | The probability you can recall the card right now, given elapsed time and stability |
These three are tuned by 17 trainable parameters (FSRS-6 added a 17th — w20 — that personalises the shape of your forgetting curve). The model is trained on your individual review history, so the algorithm progressively learns your personal memory profile.
The empirical case — why it beats SM-2
Benchmarks across ~700 million reviews from approximately 20,000 volunteer Anki users (the FSRS-6 training corpus) consistently show FSRS reduces required review count by 20-30% for the same retention target. Independent community analyses (some of which run on millions of reviews each) reach the same conclusion: FSRS produces more accurate interval predictions, meaning fewer cards are shown too early (wasted time) or too late (forgotten).
For a UPSC aspirant carrying ~3,400 cards across a 12-month cycle:
| Metric | SM-2 | FSRS-6 |
|---|---|---|
| Daily reviews (avg) | ~220 | ~155 |
| Daily time | 35-40 min | 25-30 min |
| Annual time | ~225 hours | ~155 hours |
| Saved per year | — | ~70-90 hours |
| 90-day delayed recall (90% target) | ~88% actual | ~91% actual |
That saved 70-90 hours is enough to take three more sectional mocks per month for the full year, or do a complete second read of Spectrum.
How to switch — UPSC-specific setup
- Update Anki to 25.07 or later (latest stable is 25.07+; FSRS-6.3.1 shipped March 2026).
- Switch the scheduler: In each deck's options, set scheduler to FSRS. New profiles default to FSRS automatically; older profiles do not switch on their own.
- Set desired retention to 0.90 (90%). This is the sweet spot for UPSC — high enough to feel solid in mocks, low enough not to drown in reviews. Setting it to 95% balloons reviews by ~60% for marginal recall gain.
- Wait until you have 1,000+ reviews logged, then click 'Optimise FSRS parameters' on each major deck. This trains FSRS on your memory pattern, not the generic prior.
- Re-optimise every 2-3 months. Your memory profile shifts during a UPSC cycle — early days are slower, last 90 days run faster.
When SM-2 still wins (the edge cases)
- Decks with <100 reviews logged. FSRS needs data; SM-2 works from day 1 with sensible defaults.
- Very small decks (<200 cards). Both algorithms perform similarly; algorithm choice does not move the needle.
- If you cap daily new cards at <5. Low-volume decks rarely accumulate enough data for FSRS optimisation to outperform SM-2's defaults.
For a serious UPSC stack of thousands of cards, none of these edge cases apply.
A worked example — Polity 'Schedules' deck after 1 year
- Cards: 200 (each Schedule + sub-points)
- Total reviews logged: ~6,800 across 12 months
- SM-2: average interval ~24 days; recall accuracy ~84% at 90-day delay
- FSRS-6 (post-optimisation): average interval ~31 days; recall accuracy ~89% at 90-day delay
The longer interval means fewer reviews; the higher accuracy means better Prelims-day recall. Both metrics improve simultaneously — which is why the algorithm has become the community default.
What 'FSRS-6' adds over FSRS-5
FSRS-6 (released May 2025; weights stabilised through v6.3 in October 2025; latest v6.3.1 in March 2026) added:
- w20 — a trainable parameter governing forgetting-curve shape, personalised per user
- Better calibration on irregular-review users — aspirants who skip 3-4 days during mocks then catch up
- Reduced 'shock interval' on lapses — when you forget a card, FSRS-6 shortens the next interval more conservatively than FSRS-5 did
For UPSC aspirants who study in bursts (mock-heavy weekends, light weekdays), the FSRS-6 calibration improvements matter — earlier versions over-penalised gap days.
The honest caveat
FSRS scheduling is only as good as the underlying cards. A poorly-written 'paragraph card' will still be a chore at any interval. The algorithm cannot fix:
- Cards that bundle 5 facts into one prompt
- Cards with ambiguous prompts ('Article 21?')
- Cards you copy-paste from coaching PDFs without understanding
Fix card quality first, then let FSRS do its job.
Mentor's note
If you are already on Anki with SM-2 and have <1,000 reviews logged, switching now is risk-free — FSRS will simply start with conservative priors. If you have a multi-year SM-2 deck with tens of thousands of reviews, switch but expect 1-2 weeks of slightly elevated review counts while FSRS calibrates to your history. Within a month, you will be doing fewer cards for higher retention. There is no longer a credible reason to stay on SM-2 for any UPSC use-case.
