Cursor adds Claude Opus 5 at half Fable 5's price
By Steven Van ·
Cursor's own coding benchmark shows Opus 5 matching Fable 5 while costing half as much, and it supports Zero Data Retention.
Cursor turned on Claude Opus 5 the day Anthropic shipped it, and the pitch is unusually concrete: on Cursor's own benchmark it performs at the frontier while costing half as much as the model it rivals. Opus 5 scores 66.7 on CursorBench at default effort, essentially tying Claude Fable 5's 66.5, but at half the price.
Frontier coding, half the cost
Opus 5 is Anthropic's new flagship, and its headline claim is roughly Fable 5-level intelligence at half the price ($5/$25 per million tokens vs Fable's $10/$50). Cursor's CursorBench number is the practical proof of that for coding specifically: if you were reaching for Fable 5 on hard agentic tasks, you can now get the same result for half the spend. For developers who run an agent all day, that's a direct, compounding saving, not a rounding error.
The Zero Data Retention advantage
Cursor called out something Fable 5 can't offer: Opus 5 is compatible with Zero Data Retention (ZDR). For teams under strict data-handling requirements, that's often the deciding factor, not raw benchmark scores. A model that matches the frontier and clears the compliance bar is exactly what enterprise engineering orgs have been waiting for, and it removes the awkward trade-off between using the best model and satisfying legal.
The effort toggle
Opus 5 also ships with a low/medium/high effort control, and Cursor's 66.7 is at default effort. That means there's headroom: push effort up for the genuinely hard problems, dial it down for routine edits to save tokens. It's the kind of knob that matters most inside an editor, where you're firing dozens of requests an hour at wildly different difficulty.
Why it matters
Anthropic is shipping models at a startling pace, four in two months, and Opus 5 roughly doubled Opus 4.8's score on the hardest agentic coding benchmark (Frontier-Bench: 43.3% vs 18.7%). Cursor being a same-day launch partner means you can feel that jump in your actual workflow immediately: pick Opus 5 from the model selector and point it at a real, multi-file task. If the CursorBench parity holds in your codebase, the half-price math makes it an easy default.
Sources: Cursor on X, Anthropic.