FSRS (Free Spaced Repetition Scheduler)

Definition:

FSRS is an open-source spaced repetition scheduling algorithm developed by Jarrett Ye beginning in 2022. It supersedes SM-2 as the preferred scheduler in Anki and is used natively in Sakubo. FSRS is grounded in Piotr Wozniak‘s two-component model of long-term memory and fitted to large-scale real user data using machine learning, producing better retention predictions than SM-2’s hand-tuned constants.

Also known as: Free Spaced Repetition Scheduler, open-spaced-repetition scheduler, FSRS-4, FSRS-5 (version names)

In-Depth Explanation

SM-2’s core limitation is that it uses fixed mathematical constants calibrated to Wozniak’s own learning in 1987. While these constants generalize reasonably, they cannot adapt to individual learners’ actual memory behavior — every new Anki user gets the same starting easiness factor, the same interval multipliers, the same reset behavior on failure. FSRS replaces these fixed constants with a model fitted to real data.

The two-component model: FSRS is built on the insight that every memory trace has two independent properties:

1. Retrievability (R): The probability of successful recall right now. R starts near 100% after a successful review and decays toward 0 as time passes. This is what the forgetting curve describes.

1. Stability (S): How slowly retrievability decays — the half-life of the memory. A freshly learned item has low stability (R drops to 90% within a day). A well-consolidated item has high stability (R barely changes over months). Each successful review increases stability: the memory becomes more resistant to forgetting.

The FSRS scheduling goal is to schedule the next review when R reaches a target threshold — typically around 90%. Rather than using a fixed multiplier like SM-2’s easiness factor, FSRS uses a set of model parameters (D, S, R, and several fitting constants) that are optimized for each user from their actual review history. A user who reviews consistently for a few weeks will have personalized parameters; a new user starts with population-average parameters that are still substantially better than SM-2’s defaults.

Practical differences from SM-2:

• FSRS does not catastrophically reset a card on failure — it reduces stability moderately, reflecting that residual memory exists after a lapse.
• FSRS parameters are re-optimized periodically from user data, so the algorithm improves with use.
• FSRS uses a “desired retention” setting (e.g., 90%) that directly targets a retention rate, rather than relying on an indirect proxy like easiness factor.
• Benchmarks on real Anki user data consistently show FSRS achieves the same retention with fewer total reviews, or higher retention with the same review volume.

FSRS is fully open-source, with implementations in Python, Rust, JavaScript, and other languages available at github.com/open-spaced-repetition, making it straightforward to integrate into any SRS platform.

Common Misconceptions

“FSRS needs months of data to work.”

FSRS ships with population-average default parameters derived from a large anonymized review dataset. Even on day one, before any personal data accumulates, these defaults produce better scheduling than SM-2’s fixed parameters for most users. The personalization improves over time, but the baseline is already strong.

“FSRS is just a better SM-2 — same idea, new formula.”

The fundamental architecture is different. SM-2 tracks a per-card easiness factor and uses it as an interval multiplier. FSRS tracks per-card stability and retrievability as separate variables, models their relationships using a fitted function, and schedules based on a direct retention probability target. These are different conceptions of memory, not just different formulas.

“Setting a high desired retention (e.g., 95%) is always better.”

Higher desired retention requires more frequent reviews — significantly more. The relationship is nonlinear: going from 90% to 95% retention roughly doubles review burden. For most learners and materials, 85–92% is the practical sweet spot — high enough for solid retention, low enough to remain sustainable. FSRS makes this tradeoff explicit and configurable, unlike SM-2 which offers no direct retention target.

“FSRS is only for Anki power users.”

FSRS has been the default scheduler recommendation in Anki since late 2022 and is the native scheduler in Sakubo. Any user who enables it and sets a retention target gets the full benefit without needing to understand the underlying algorithm. The technical complexity is encapsulated in the tool.

History

• 2022: Jarrett Ye releases the first version of FSRS through the open-spaced-repetition GitHub project. The algorithm is grounded in Wozniak’s two-component memory model and trained on real Anki review data to derive fitted parameters. [Ye, 2022]

• 2022–2023: FSRS goes through versions 4, 4.5, and 5, each incorporating more user data and refined model parameters. Benchmarks comparing FSRS and SM-2 on held-out review data demonstrate statistically significant improvements in retention prediction accuracy.

• 2022–present: Damien Elmes integrates FSRS into Anki as an optional then recommended scheduler. This places FSRS in front of Anki’s multi-million user base — the largest deployment of a new SRS algorithm since SM-2 was implemented in 2006.

• 2023–present: FSRS is adopted by Sakubo, Mochi, and other SRS platforms. The open-spaced-repetition project provides implementations in multiple languages for easy integration. Community benchmarking and algorithm refinement continue actively.

Criticisms

FSRS, while showing superior empirical performance over SM-2 in benchmark tests, has been criticized for the black-box nature of its machine-learning-derived parameters — the algorithm optimizes for minimum review count and maximum retention, but the relationship between its internal parameters and explicit theories of memory consolidation is less transparent than rule-based SM-2. Some SRS practitioners value algorithmic legibility for calibrating their own review behavior; FSRS’s complexity makes manual parameter adjustment less intuitive. As a newer algorithm, its long-term performance data (particularly for retention across 5-10+ year timeframes) is more limited than SM-2’s decades of community use.

Social Media Sentiment

FSRS is enthusiastically received in the Anki community following its integration as the default scheduling option in Anki 23.10. Benchmark comparisons showing FSRS achieving target retention rates with significantly fewer reviews than SM-2 are widely shared and discussed. The algorithm’s creator (Jarrett Ye / L-M-Sherlock) has been active in the community, providing transparency about methodology and benchmarking. Anki power users consistently report that switching to FSRS reduces daily review burden while maintaining retention levels — concrete measurable results that drive strong community adoption and advocacy.

Last updated: 2026-04

Practical Application

FSRS is the currently recommended scheduling algorithm for Anki users — enabling it via Anki’s settings reduces daily review workload at the same retention level compared to SM-2. The desired retention slider allows learners to calibrate the frequency-retention tradeoff: higher settings (90%+) reduce forgetting but increase review load; lower settings (80%) reduce workload at the cost of slightly more relearning. Sakubo applies optimized spaced repetition scheduling for vocabulary review, capturing the core FSRS benefit — maximally efficient review intervals — in a contextual vocabulary learning environment without requiring manual algorithm configuration.

Related Terms

• SRS (Spaced Repetition System)
• SM-2 (SuperMemo 2)
• Forgetting curve
• Long-term memory
• Anki

See Also

• Jarrett Ye
• Piotr Wozniak
• Study queue
• Retrieval practice

Research

• Ye, J. et al. (2022–present). FSRS algorithm repository. https://github.com/open-spaced-repetition/fsrs4anki
  Summary: The primary source — full documentation of the FSRS algorithm including model architecture, parameter derivation, and implementation details. The most authoritative reference for understanding FSRS’s design and deployment.

• Ye, J. (2023). FSRS vs SM-2 benchmark. https://github.com/open-spaced-repetition/fsrs-vs-sm2
  Summary: Systematic empirical comparison on real Anki review data. Demonstrates FSRS’s statistically significant improvement in retention prediction over SM-2. The key evidence base for FSRS adoption.

• Wozniak, P.A. (1995). Two components of long-term memory. supermemo.com. https://www.supermemo.com/en/blog/two-components-of-long-term-memory
  Summary: The two-component model (retrievability and stability) that FSRS formalizes computationally. Wozniak proposed this model theoretically; FSRS implements it in a machine learning framework fitted to large-scale data.

• Wang, Y. (2022). A stochastic shortest path algorithm for optimizing spaced repetition scheduling. arXiv preprint. https://arxiv.org/abs/2207.14255
  Summary: Academic treatment of the optimization problem behind modern SRS scheduling, including the mathematical framework that informs FSRS’s approach to minimizing reviews while achieving a retention target.

• Anki documentation — FSRS. https://docs.ankiweb.net/deck-options.html#fsrs
  Summary: Official Anki documentation for the FSRS scheduler — explains the desired retention setting, parameter optimization workflow, and practical guidance for users transitioning from SM-2. The clearest user-facing documentation of FSRS in practice.