How to Use Claude Fable5 in Cursor & Claude Code (Setup Guide)

Update — July 25, 2026: Fable 5 remains live in Claude Code. From July 20, Max and Team Premium permanently include Fable 5 for up to 50% of weekly limits; Pro typically needs usage credits. If the model picker will not show Fable, follow Why can't I select Fable 5?. Safety classifiers may still route some coding turns to Opus 4.8. Live status: /status · Plans: /claude-fable-5-max-plan.

Claude Fable5 landed in both Cursor and Claude Code on launch day, and it tops SWE-Bench Pro at 80.3% — 11 points ahead of Opus 4.8. Here is exactly how to switch each tool over, what changes in your workflow, and what it costs when you do.

Use Claude Fable5 in Cursor

1. Open the model settings. Cmd/Ctrl + Shift + J → Models (or Settings → Cursor Settings → Models).
2. Enable Claude Fable5. Find claude-fable-5 in the model list and toggle it on. If you bring your own Anthropic API key, paste it under API Keys first — Fable5 requires a key with access to the model (any funded Claude Console account has it).
3. Select it in chat or agent mode. Click the model name in the composer's model picker and choose Claude Fable5. Cursor also exposes effort variants (e.g. high-effort thinking) — higher effort means deeper reasoning, more tokens, slower responses.
4. Scope it to hard tasks. A practical setup: keep your default model on something cheaper, and switch to Fable5 per-conversation for long agent runs and gnarly debugging.

That's it — there is no Fable5-specific configuration beyond the picker. One behavioral note: Fable5 follows instructions more literally than older Claude models, so terse, precise prompts work better than long hedged ones. Our prompt guide covers the details.

Use Claude Fable5 in Claude Code

Three ways, from quickest to most permanent:

# 1. Inside a running session — switch on the fly
/model claude-fable-5

# 2. Launch a session with Fable5
claude --model claude-fable-5

# 3. Make it the default (settings.json)
{ "model": "claude-fable-5" }

You can also set the environment variable once and forget it:

export ANTHROPIC_MODEL=claude-fable-5

Since July 24, 2026, Claude Opus 5 is the default model on Max and the strongest model available on Pro, so Fable5 is now an explicit choice rather than what you get by default. To go the other way — /model claude-opus-5 in a session, or export ANTHROPIC_MODEL=claude-opus-5 — costs $5/$25 instead of $10/$50.

Subscription users: from July 20, 2026, Max and Team Premium permanently include Fable5 for up to 50% of weekly usage limits; Pro typically bills via usage credits. API-key users pay standard API rates. Plan matrix: Is Fable 5 included in Claude Max?. Cost math: pricing calculator.

Running on AWS? Point Claude Code at Bedrock instead with CLAUDE_CODE_USE_BEDROCK=1 — but note Bedrock is still being re-enabled region by region after the restoration, and access may be gated. See Claude Fable 5 on AWS Bedrock for the identity-verification and use-case approval steps.

What changes in an agentic coding workflow

• Longer leashes work. Fable5 sustains multi-hour agent runs and keeps its own progress notes. Give it the full task spec up front instead of drip-feeding steps.
• No temperature, no top_p. Sampling knobs are gone; the model is steered entirely through prompts and the effort parameter (low → max). For coding agents, high is the sweet spot; reserve max for frontier-hard problems.
• Safeguard fallback exists in IDEs too. Security-adjacent work (exploit analysis, malware-ish code) may be silently answered by Opus 4.8 — you'll see the same behavior in Cursor and Claude Code as in the API. Since the July 1 restoration, the new safety classifiers make this slightly more common even on ordinary coding/debugging requests while Anthropic tunes false positives. How the fallback works.

What a coding month actually costs

The sticker price ($10/$50 per million tokens) is double Opus 4.8 — but agent coding is cache-heavy, and cached input is 90% off. Realistic numbers for a heavy user:

The escalation pattern is the one to copy: route 10–20% of frontier-hard tasks to Fable5 and the premium costs less than a Cursor seat. On the hardest problems the math flips entirely — one avoided retry pays the whole 2x. Run your own numbers in the cost calculator.

FAQ

Is Claude Fable5 available in Cursor? Yes, since June 9, 2026. Enable it under Settings → Models and pick it from the model picker.

How do I switch Claude Code to Fable5? /model claude-fable-5 inside a session, or claude --model claude-fable-5 at launch.

Is Fable5 worth 2x the price for coding? For long agentic runs and hardest-tier bugs, yes — community cost-per-solve estimates put it cheaper than Opus 4.8 there. For routine edits, no — keep a cheaper default.

Does the July 7 billing deadline affect Cursor users? Only indirectly. The July 7 cutoff (which replaced the original June 22 window after the restoration) applies to Claude subscription plans (claude.ai apps and Claude Code on Pro/Max/Team). Cursor usage bills through Cursor's own pricing or your API key.

Model behavior and pricing from Anthropic's June 9, 2026 announcement. See the full Fable5 guide and prompt generator.