How to Use Claude Fable 5 in Cursor (2026 Step-by-Step)

Claude Fable 5 landed in Cursor on June 9, 2026 and immediately set a new record on CursorBench at 72.9%, beating the previous best by 8 points. It is Anthropic's most capable model for autonomous coding and long-running agentic work.

But Fable 5 is not a model you want to use for every request. At $10/$50 per million tokens, it costs about twice as much as Claude Opus 4.8. One user reported spending $1.72 on a small bug fix that would have cost 13 cents with Composer 2.5.

This guide walks through the exact setup, the cost math you need to understand before starting, and when to reach for Fable 5 versus a cheaper model. If you want a broader comparison of Fable 5 against other frontier models, see our Claude Fable 5 vs GPT-5.6 Sol comparison.

What you'll need

• Cursor installed and updated. Custom model support requires a recent version. If you have not updated in a while, do that first.
• A paid Cursor plan. Bring-your-own-key and custom model selection require at least Cursor Pro ($20/month). The free tier only gives you Auto mode, which does not let you pin a specific model.
• An Anthropic API key with billing enabled. Create a key from the Anthropic developer console. Your account needs a funded workspace, not just a login. Fable 5 bills directly to your Anthropic account.

Step 1: Add your Anthropic API key to Cursor

Open Cursor and head to Settings > Models. You can get there from the gear icon or the command palette.

Scroll down to the provider keys section. You will see labeled fields for Anthropic, OpenAI, and Google. Find the Anthropic API key field, paste your key, and click Verify.

Cursor sends a test request to confirm the key works and your account can reach the Anthropic API. A green checkmark means you are good. If verification fails, the usual causes are a typo, a key from the wrong workspace, or an Anthropic account with no billing set up.

Once verified, toggle the Anthropic key on so Cursor actually routes Claude requests through your key instead of its bundled credits.

Step 2: Enable Fable 5 as a model

In the Models list, look for the add or search model box. Type claude-fable-5 exactly. When it appears in the results, click Add to add it to your model list.

After adding it, confirm the toggle next to claude-fable-5 is on. A disabled entry will not show up in the model picker when you are working in a file. This is the most common reason a freshly added model seems to vanish.

Step 3: Handle Privacy Mode and data retention

Fable 5 has special privacy requirements because of how Anthropic handles data for Mythos-class models. Whether you need to take action depends on your account type:

• Individual accounts without Privacy Mode: Fable 5 is on by default. No action needed.
• Individual accounts or Teams with Privacy Mode enabled: Fable 5 is off by default. You must approve the Data Retention Policy in the Cursor Dashboard before you can use it.
• Enterprise accounts: Fable 5 is off by default regardless of Privacy Mode. An admin must enable it at the Organization level, and access can be restricted per team or group.

When you enable Fable 5, Anthropic stores agent input and output data for automated and human harm-prevention processes. This data is deleted after 30 days, is not readable by people by default, and is not used to train Anthropic's models.

Step 4: Select Fable 5 and verify it works

Adding a model and using it are separate steps. Cursor lets you pick a model per conversation.

Open a chat or the agent panel. Find the model selector near the input box or in the chat header. Click it and choose claude-fable-5 from the list.

The selection sticks for the current conversation until you change it. You can run one thread on Fable 5 and another on a cheaper model side by side.

To verify it is working, send a small prompt like "summarize the current file." If Fable 5 is active, you will see it respond. If you get an error, check the troubleshooting section below.

Step 5: Use the deep reasoning variant when needed

Fable 5 supports a thinking variant for problems that benefit from deeper deliberation. Cursor recommends the high thinking variant for the strongest results.

When you select Fable 5 in the model picker, look for the thinking or reasoning effort option. Set it to high for complex architectural decisions, multi-file refactors, or debugging sessions where the model needs to reason through interlocking dependencies.

For simpler tasks, skip the thinking variant. It burns more tokens and takes longer without adding value on straightforward requests.

How much Fable 5 costs in Cursor

Fable 5 draws from Cursor's API pool, which charges at the model's standard API rates. Here is the pricing:

For comparison, here is how Fable 5 stacks against other Cursor models:

A typical agent run that processes 50K input tokens and generates 10K output tokens costs about $1 with Fable 5. The same task with Composer 2.5 costs about 5 cents.

Cursor plans include monthly API credits that offset these costs:

At Fable 5's rates, the Ultra plan's $400 API allowance covers roughly 13 million input tokens or about 8 million output tokens per month. That is enough for a few dozen long agent runs if you are selective about when you use Fable 5.

When to use Fable 5 (and when not to)

Fable 5 was built for long-horizon, autonomous work. It holds context across millions of tokens and drives tasks to completion with fewer check-ins than any previous Claude model.

Use Fable 5 for:

• Multi-file refactors that span dozens of files
• Framework migrations or ORM upgrades across a codebase
• Large API surface changes that must stay consistent everywhere
• Untangling legacy modules where the model needs to hold interlocking context
• Long agent runs where you want to walk away and let it work

Do not use Fable 5 for:

• One-line fixes or quick tests
• Small component creation
• Simple renaming tasks
• Anything where Composer 2.5 or Auto would handle it in seconds

The skill is matching the model to the size of the job. Use Fable 5 for the hardest 20% of your work and switch to a cheaper model for the other 80%.

Common errors and fixes

Error: Fable 5 does not appear in the model picker Make sure the toggle next to claude-fable-5 is on in Settings > Models. If you just added it, restart the chat panel or reopen the file. If it still does not show, verify your Anthropic key is enabled and verified.

Error: "routing_unsupported" or "requires non-ZDR provider policy" This error appeared for some users around June 12, 2026. It happens when your Anthropic account uses Zero Data Retention (ZDR) mode, which Fable 5 does not support. To fix it, switch your Anthropic account to 30-day data retention mode in the Anthropic console. If you cannot change your retention policy, call the Anthropic API directly outside Cursor as a workaround.

Error: Verification fails when adding the API key Check that the key is from the correct Anthropic workspace, that billing is enabled, and that there are no typos. Keys from a workspace without billing will verify as invalid.

Error: Costs are higher than expected Fable 5's thinking variant uses significantly more tokens. If you are seeing unexpectedly high costs, check whether the thinking variant is enabled and switch to the standard variant for routine tasks. Also check your Cursor usage dashboard at cursor.com/dashboard/usage to see which requests are consuming the most tokens.

What to build next

Now that Fable 5 is set up, here are some ways to put it to work:

• Try it on a multi-file refactor you have been putting off. Give the agent a clear goal and let it run.
• Compare Fable 5 against GPT-5.6 Sol on the same task. Our Fable 5 vs GPT-5.6 Sol comparison breaks down which model wins for different task types.
• Learn how to vibe code with Lovable and Cursor to round out your AI coding workflow.
• Explore the best AI app builders for vibe coding to see how Cursor fits into the broader landscape.

If you want to go deeper on AI coding tools, the AI SaaS Course covers building complete applications with AI-assisted development.