FSRS: A modern, efficient spaced repetition algorithm
A new spaced repetition algorithm called FSRS is attracting interest from Anki and language‑learning enthusiasts, who report needing 20–30% fewer reviews compared to classic SM‑2-based scheduling. Commenters highlight that while FSRS and similar models can adapt to deck content and individual performance, the biggest gains often come from how flashcards are designed—favoring easier, atomic, context-rich cards over dense or abstract ones. Many see SRS as just one component of effective learning, arguing it works best when combined with immersion, better tooling for card creation, and varied, real‑world practice rather than rote memorization alone.
Perceived advantages of FSRS over classic algorithms
- Benchmarks cited claim 20–30% fewer reviews compared to classic Anki/SM‑2 while maintaining target retention.
- Users report it removes much of the manual tweaking needed with SM‑2, especially for decks with different optimal retention levels.
- FSRS is said to handle late reviews gracefully by updating stability based on actual recall rather than punishing lateness.
Implementations, ecosystem, and integration
- Multiple open-source libraries exist (JS, Go, Rust, Python, Clojure, Swift, Elixir), all MIT-licensed, making FSRS easy to embed in custom apps.
- FSRS is integrated into Anki (with retraining/recompute features) and is stable on iOS Anki, available in AnkiDroid alpha builds.
- Some developers are considering switching their own SRS tools from SM‑2 to FSRS.
Experiences and benchmarking skepticism
- Heavy users report better experience and less “ease hell,” though benefits are said to appear mainly after thousands of reviews.
- Benchmarks vs SM‑17 are shared; some participants question statistical significance and call for more rigorous analysis and peer-reviewed evidence.
- There is interest in comparing FSRS to SuperMemo’s newer SM‑18, but current tooling/APIs make that hard.
Card design and learning strategy matter more than algorithms
- Many argue the biggest gains come from better card formulation, not marginal algorithm improvements.
- Strong emphasis on: making cards easier/atomic, aggressively deleting or suspending bad cards, using cloze deletions (including partial-word clozes), and minimizing repetitions.
- References to SuperMemo’s “20 rules” and other guides on writing good prompts.
- Consensus that overwhelming decks and complicated cards often cause people to quit SRS entirely.
Language learning use cases and tools
- Extensive discussion of sentence/cloze cards, context-rich cards, and “lazily graded” passes to keep momentum.
- Several tools and startups are mentioned: automatic sentence mining, YouTube-to-Anki pipelines, ebook/subtitle importers, and systems that combine FSRS-like scheduling with continuous dosing of new cards instead of strict due dates.
- Diverging views on Duolingo vs SRS: some see cards as far more effective, others argue they serve different roles and should be combined.
Non-language applications and alternative models
- Uses include guitar practice prompts, music theory, math proofs, geography, song lyrics, programming concepts, and trivia (with elaborate mnemonic image systems).
- Some want algorithms without hard due dates, using probability-based sampling instead; others argue that without scheduling aligned to forgetting curves, effectiveness drops.
- Modular architectures separating cards, history, and “just-in-time” schedulers are discussed as a desirable direction.