Verified September 2, 2026
Claude Fable 5.1 in Cursor
Selecting the model takes ten seconds. Understanding which turns bill at $50 per million output — and when Cursor routes you away from Fable 5.1 — is the actual work.
The short answer
Supported on launch day. Open Settings → Models, pick claude-fable-5-1 (not the legacy claude-fable-5 entry if both appear), and confirm the model badge at session start. Use Max Mode only when the task needs the full 1M window — it bills at the API rate. BYOK users paste an Anthropic key under Models; bundled-plan users draw from fast-request quota first, then usage-based billing.
Support status
Fully supported, verified September 2, 2026. Cursor's changelog lists claude-fable-5-1 alongside the September Anthropic release. It sits at the top tier of Cursor's model routing for Agent and Composer. The model ID is claude-fable-5-1 — hyphenated, no date suffix in the picker. Cursor documents the same caveats Anthropic does: mandatory 30-day retention, no temperature parameter, and safety-classifier refusals that can route a turn to a fallback model without a loud error.
Setup
Five steps — the last two are the ones that save money.
- 1
Update Cursor
Fable 5.1 requires a current Cursor build — the model picker is server-driven, so an old client may still show claude-fable-5 without the 5.1 entry. Check for updates before you assume the model is missing.
- 2
Select claude-fable-5-1 in Models
Open Cursor Settings → Models and enable claude-fable-5-1 for Agent and/or Composer. If both claude-fable-5 and claude-fable-5-1 appear, pick 5.1 — the older ID lacks the September reasoning upgrades and higher SWE-bench score.
- 3
Choose bundled quota or BYOK
Pro and Business plans include fast requests that consume quota before usage billing kicks in. BYOK users add an Anthropic API key under Models — you pay Anthropic directly and see usage in the Anthropic console. The model behaves the same; only the invoice and rate limits differ.
- 4
Use Max Mode deliberately
Max Mode unlocks the full 1M context window and bills at API rates ($10/$50 per million tokens as of September 2026). For everyday edits, leave it off — Agent mode with a smaller context slice is cheaper and often sufficient.
- 5
Verify the model badge each session
Cursor can silently route a flagged turn to a fallback model. The model indicator in the chat panel is the only reliable proof you are on Fable 5.1. If routing matters for a benchmark or cost attribution, check the badge at session start and after any refusal.
Gotchas
Four things that surprise people in week one.
claude-fable-5 and claude-fable-5-1 are different models
Cursor may list both during the transition. The 5.1 ID carries the September release — higher SWE-bench, reduced latency, adaptive thinking budget. Picking the legacy claude-fable-5 entry means you are not on 5.1 even if the UI label looks similar.
No temperature — effort is the quality lever
Fable 5.1 rejects temperature, topP, and topK. Cursor passes effort through its model settings when exposed; if your workflow relied on temperature for determinism, switch to explicit output-format instructions instead.
Agent loops resend context every turn
Cursor's Agent mode resends file context on every step. With a 1M window nothing forces truncation, so a long task quietly grows per-turn input until each step costs several dollars in input alone. Break work into checkpoints and start fresh threads rather than letting one session run all afternoon.
Refusals can route without a clear error
When the safety classifier fires, Cursor may serve the turn from a fallback model — the chat continues, but you are no longer on Fable 5.1. Watch the model badge. Production workflows should define which model picks up after a refusal.
Cost notes
Fable 5.1 is $10 per million input and $50 per million output at API rates (September 2026). Bundled fast requests cover a portion of usage on Pro/Business; beyond quota, usage billing applies at the same rates. BYOK users get Anthropic prompt caching (90% discount on cache hits) — keep the model stable mid-task so the cache stays warm. Max Mode is the fastest way to burn through quota on a large repo.
Know what a turn costs before you run it
Our cost calculator models Fable 5.1 input/output/cache math for Agent-shaped workloads — the pattern Cursor's loops actually generate.