Claude Fable 5.2 lands in Cursor and tops CursorBench at 60%

TL;DR: Claude Fable 5.2, Anthropic's newest Mythos-class model, is now available in Cursor. Cursor says it is the top model on CursorBench at 60% on max effort, and it scores 56.9% at the default high effort, up from 49.2% for Fable 5.1. Token prices are 20% lower than Fable 5.1, but a typical benchmark task still costs a bit more.

What Cursor announced

On November 9, 2026, Cursor posted on X that "Claude Fable 5.2 is now available in Cursor" and that "at max effort, it's the top model on CursorBench at 60%" (Cursor on X). Cursor's model docs say Fable 5.2 replaces Fable 5.1 and is built to plan, hand work to subagents, and keep going on long, ambiguous, multi-step problems (Cursor Docs).

The CursorBench numbers

CursorBench is Cursor's own test set, built from ambiguous, multi-file tasks taken from real Cursor sessions. Here is how Fable 5.2 lands on the live CursorBench 4.0 table (CursorBench):

| Model and effort | Score | Cost per task | Steps per task | | --- | --- | --- | --- | | Fable 5.2 Max | 60.0% | $32.79 | 218 | | Opus 5.5 Max | 57.8% | $13.43 | 185 | | Fable 5.2 High (default) | 56.9% | $9.70 | 69 | | Opus 5.5 High | 56.0% | $3.97 | 68 | | Fable 5.1 High | 49.2% | $9.08 | 77 |

Two things stand out. The default setting jumped 7.7 points over Fable 5.1 while taking fewer steps. And max effort buys about three more points at more than three times the cost per task, so it is worth saving for the hardest jobs.

Pricing

Cursor lists Fable 5.2 at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 on Fable 5.1 (Cursor Docs). Even so, the default-effort run on CursorBench costs $9.70 per task versus $9.08 for Fable 5.1, because the new model uses more tokens per task. Cursor also notes its rates are roughly twice Claude Opus 5.5's. It is a third-party model, so it bills from prepaid credit rather than included plan usage.

How to use it well

• Start at high. Cursor says Fable 5.2's effort levels behave differently from Fable 5.1's, so don't reuse your old setting. Extra high and max are for long, hard tasks.
• Check your privacy settings. Fable 5.2 is on by default for personal and Team accounts with Privacy Mode off. It is off by default for Teams with Privacy Mode on and for every Enterprise customer, and an admin has to accept Anthropic's data-retention terms first.
• Expect an automatic fallback. If a request trips one of Fable 5.2's safety guardrails, Cursor sends it to Claude Opus instead, so the task keeps going without an error.

Why it matters

Cursor is one of the main places developers try new coding models side by side, and it shipped Fable 5.2 the day Anthropic announced it (Anthropic). Keep in mind that CursorBench is Cursor's own benchmark, so rankings may look different in other tools and on other codebases.