FSRS vs SM-2 — how much better is it, and why?

On the open-spaced-repetition benchmark (hundreds of millions of real Anki reviews) FSRS predicts recall far more accurately — roughly ±5% error vs SM-2's ±16% — which cashes out as about 20–30% fewer reviews for the same 90% retention.

Why it wins:

• it models forgetting instead of using one mechanical multiplier,
• it tracks difficulty separately from stability (no single "ease factor" to corrupt),
• it handles missed or early reviews gracefully instead of resetting the card, and
• it lets you dial retention explicitly to trade time against forgetting.

Same effort budget, more remembered — or same retention, less time. That's the practical difference.