How do I switch Claude Code to Claude Fable 5?

In Claude Code, set the default model in ~/.claude/config.json by setting the model field to 'claude-fable-5'. Or use the --model flag per session: 'claude --model claude-fable-5'. As of June 9, 2026, claude.ai Pro and Max subscribers see Fable 5 in the model picker and Claude Code defaults to it for those tiers. Free users still default to Haiku 4.5 and need to upgrade for Fable 5 access.

How do I enable Claude Fable 5 in Cursor 4?

In Cursor 4, open Settings → Models, find Claude Fable 5 in the list, and either set it as the default or pin it to specific agents. If using bring-your-own-API-key mode, ensure your Anthropic API key has access (all Anthropic API users get Fable 5 access by default as of June 9, 2026). For team workspaces, the org admin can set Fable 5 as the team default for specific projects. Note that Cursor will pass through 2x the API cost of Opus 4.8 when running Fable 5.

Will Claude Fable 5 break my existing prompts?

Mostly compatible, but with two gotchas. First, the new safeguard layer refuses some queries that Opus 4.8 would handle — primarily offensive security, biology dual-use, and certain CTF/red-team workflows. Review your prompts and remove any phrasing that resembles offensive use. Second, Fable 5 reasons differently — more deeply, with longer chain-of-thought — so over-specified prompts that worked for Opus 4.8 may produce more verbose outputs. Plan for a one-week prompt re-tuning window for production systems.

How do I roll back from Claude Fable 5 if needed?

Easy in all environments. Claude Code: 'claude --model claude-opus-4-8' per session or edit ~/.claude/config.json. Cursor 4: Settings → Models → set default to Claude Opus 4.8. claude.ai: open the model picker in any conversation and select Opus 4.8. Bedrock/Vertex: change the model ID in your API call from claude-fable-5 to claude-opus-4-8. Anthropic continues to support Opus 4.8 indefinitely — there is no forced migration to Fable 5.

Quick Answer

How to Switch to Claude Fable 5 in Claude Code and Cursor

Published: June 10, 2026

How to Switch to Claude Fable 5 in Claude Code and Cursor 4

Claude Fable 5 launched June 9, 2026 as Anthropic’s first publicly available Mythos-class model. Here’s how to enable it across every common entry point — and when to stay on Opus 4.8.

Last verified: June 10, 2026

Before you switch — should you?

Quick gut check (fuller breakdown):

Yes, switch if: you run long-horizon autonomous coding agents, hard reasoning tasks, or research-grade work
Don’t switch yet if: you do high-volume routine code edits, RAG, classification, chat — Opus 4.8 is half the price
Hybrid is best: Opus 4.8 default + Fable 5 for hard tasks

1. Switch in Claude Code

Set as default

Edit ~/.claude/config.json:

{
  "model": "claude-fable-5",
  "fallback_models": ["claude-opus-4-8", "claude-haiku-4-5"]
}

Per-session

claude --model claude-fable-5

Per-task in a script

claude --model claude-fable-5 --skill refactor "Refactor src/api/* to use the new schema"

Pro/Max subscribers (claude.ai)

As of June 9, 2026, Claude Code Pro and Max tiers default to Fable 5 when available, falling back to Opus 4.8 when credits are tight. Verify with:

claude config show

Cost warning

Fable 5 is 2x the API cost of Opus 4.8. If you run Claude Code with a non-subscription API key, expect bills to roughly double overnight on migrated workloads.

2. Switch in Cursor 4

Set as default

Open Cursor 4 → Settings (⌘,)
Models → Provider: Anthropic
Default model: Claude Fable 5
Save

Per-task in agent mode

Open the agent panel (⌘L)
Click the model picker (top right of agent panel)
Select Claude Fable 5
Run

Pin per project

In Cursor 4 team workspaces:

Workspace settings → Defaults → Per-agent model
Pin “Refactor agent” to Fable 5
Pin “Inline completion” to Cursor Tab or Opus 4.8 (cheaper for fast UX)

Bring-your-own-API-key

If using BYOK mode (passes API cost directly to your Anthropic billing):

Settings → API Keys → Anthropic → paste key
Confirm Fable 5 is listed as available
Monitor cost via Anthropic Console

3. Switch in claude.ai (web)

Pro / Team / Enterprise subscribers

Open any conversation on claude.ai
Click the model name at the top of the conversation
Select Claude Fable 5
Fable 5 will be used for that conversation; new conversations follow the default

Set Fable 5 as your account default

Account Settings → Preferences → Default model → Claude Fable 5

Free tier

Free claude.ai users do not get Fable 5. Upgrade to Pro ($20/mo) for access. Note that the June 22, 2026 usage-credits change makes Fable 5 consume credits roughly 2x faster than Opus 4.8 from your Pro quota.

4. Switch in the API

REST

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-fable-5",
    "max_tokens": 4096,
    "messages": [
      {"role": "user", "content": "Refactor this function for clarity."}
    ]
  }'

Python SDK

import anthropic

client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-fable-5",
    max_tokens=4096,
    messages=[
        {"role": "user", "content": "Refactor this function for clarity."}
    ]
)

TypeScript SDK

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic();
const response = await client.messages.create({
  model: "claude-fable-5",
  max_tokens: 4096,
  messages: [{ role: "user", content: "Refactor this function for clarity." }],
});

5. Switch in Amazon Bedrock

Model ID

anthropic.claude-fable-5

Boto3

import boto3, json
runtime = boto3.client("bedrock-runtime")
response = runtime.invoke_model(
    modelId="anthropic.claude-fable-5",
    body=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 4096,
        "messages": [{"role": "user", "content": "Hello"}],
    }),
)

Region availability

Bedrock Fable 5 is available in us-east-1, us-west-2, and eu-central-1 at launch. Other regions follow over 2–4 weeks.

6. Switch in Google Vertex AI

Model ID

claude-fable-5

Python SDK

from anthropic import AnthropicVertex

client = AnthropicVertex(region="us-east5", project_id="your-project")
response = client.messages.create(
    model="claude-fable-5",
    max_tokens=4096,
    messages=[{"role": "user", "content": "Hello"}],
)

Region availability

Vertex AI Fable 5 launches in us-east5, europe-west1, asia-southeast1. Check Anthropic Vertex docs for current list.

7. Switch in OpenClaw / agent runtimes

OpenClaw and similar agent runtimes that proxy to Anthropic typically need a config update:

# openclaw config
agents:
  default:
    provider: anthropic
    model: claude-fable-5
    fallback: claude-opus-4-8

Restart the runtime to pick up the change.

Common gotchas

1. Cybersecurity false positives

The Fable 5 safeguard layer refuses queries that Opus 4.8 would handle. Common false positives:

CTF challenge writing
Penetration test report drafts
Malware analysis education
Some infrastructure-as-code patterns

Workaround: re-phrase the prompt to emphasize defensive intent, or fall back to Opus 4.8 for that task. See: Claude Fable 5 cybersecurity restrictions explained.

2. Token-cost shock

Fable 5 is $10/$50 vs Opus 4.8 $5/$25 per MTok — exactly 2x. Your monthly Anthropic bill will roughly double on migrated workloads. Re-baseline cost dashboards.

3. June 22, 2026 usage-credits change

claude.ai Pro/Max subscriptions get a new credits system on June 22. Fable 5 consumes credits faster than Opus 4.8. Heavy users may exhaust monthly Pro quotas sooner than expected.

4. Latency

Fable 5 reasons more deeply per token, adding 10–30% wall-clock time per response in interactive use. For chat UX and inline code completion, you may prefer Opus 4.8 or Cursor Tab for snappiness.

5. Prompt re-tuning

Old prompts optimized for Opus 4.8 may produce more verbose outputs on Fable 5. Plan a one-week tuning window for production prompt libraries.

6. MCP tool compatibility

Fable 5’s MCP integration is identical to Opus 4.8 — no tool migration needed.

Recommended rollout

Day 1

Switch personal Claude Code default to Fable 5
Test on 5 representative tasks
Note any refusals or quality regressions

Day 2–3

Enable Fable 5 for one team / one project
Monitor cost dashboards and refusal logs
Re-tune any prompts that produce verbose or refused outputs

Day 4–7

Roll out to full team or full project
Set fallback to Opus 4.8 in config
Update cost forecasts

Day 8+

Decide which workloads stay on Fable 5 vs fall back to Opus 4.8
Consider orchestrator-Opus + subagent-Haiku for high-volume work

Rollback (if needed)

Environment	Rollback command
Claude Code	`claude --model claude-opus-4-8`
~/.claude/config.json	Change `model` to `claude-opus-4-8`
Cursor 4	Settings → Models → Default → Claude Opus 4.8
claude.ai	Model picker → Claude Opus 4.8
API	Change `model` field to `claude-opus-4-8`
Bedrock	Use `anthropic.claude-opus-4-8`
Vertex AI	Use `claude-opus-4-8`

Anthropic continues to support Opus 4.8 indefinitely. No forced migration.

Sources

Anthropic Newsroom: Claude Fable 5 and Claude Mythos 5 (June 9, 2026)
Anthropic API documentation
agentpedia.codes: Claude Fable 5 Benchmarks and Prompting Guide
llm-stats.com: Claude Fable 5 Review
TrueFoundry: Claude Fable 5 API Benchmarks Pricing
claudefa.st: Claude Fable 5 Benchmarks, Pricing & June 22 Catch
AWS Bedrock documentation
Google Cloud Vertex AI Anthropic models

How to Switch to Claude Fable 5 in Claude Code and Cursor 4

Before you switch — should you?

1. Switch in Claude Code

Set as default

Per-session

Per-task in a script

Pro/Max subscribers (claude.ai)

Cost warning

2. Switch in Cursor 4

Set as default

Per-task in agent mode

Pin per project

Bring-your-own-API-key

3. Switch in claude.ai (web)

Pro / Team / Enterprise subscribers

Set Fable 5 as your account default

Free tier

4. Switch in the API

REST

Python SDK

TypeScript SDK

5. Switch in Amazon Bedrock

Model ID

Boto3

Region availability

6. Switch in Google Vertex AI

Model ID

Python SDK

Region availability

7. Switch in OpenClaw / agent runtimes

Common gotchas

1. Cybersecurity false positives

2. Token-cost shock

3. June 22, 2026 usage-credits change

4. Latency

5. Prompt re-tuning

6. MCP tool compatibility

Recommended rollout

Day 1

Day 2–3

Day 4–7

Day 8+

Rollback (if needed)

Related reading

Sources