Claude Fable 5 — Anthropic’s first generally available Mythos-class model — is rolling out inside GitHub Copilot as of June 9, 2026. It is built for long-horizon, autonomous coding, and GitHub’s own internal benchmarks found it completing equivalent autonomous coding work with fewer tool calls and lower token consumption than previous Opus-tier models.
Quick answer: Claude Fable 5 is available to Copilot Pro+, Max, Business, and Enterprise users via the model picker. Business and Enterprise admins must enable the Claude Fable 5 model policy first — until they do, the model won’t appear for their developers. It is billed at provider list pricing under usage-based billing, not as a standard premium-request multiplier. And unlike every other Claude model in Copilot, Fable 5 requires 30-day data retention so Anthropic can run its safety classifiers.
This guide covers Copilot specifics. For pricing, API access, benchmarks, and safeguards in full, see the main Claude Fable 5 guide.
Which Copilot Plans Get Claude Fable 5
| Copilot plan | Claude Fable 5 access |
|---|---|
| Copilot Free | Not available |
| Copilot Pro | Not available |
| Copilot Pro+ | Available |
| Copilot Max | Available |
| Copilot Business | Available after admin enables the model policy |
| Copilot Enterprise | Available after admin enables the model policy |
The rollout is gradual, so the model may not appear in every account immediately even on an eligible plan. If you are on Pro+ or Max and don’t see it yet, wait for the rollout rather than assuming a configuration problem.
Supported Surfaces and IDEs
Once enabled, Claude Fable 5 shows up in the model picker across Copilot’s surfaces:
- Visual Studio Code — all modes: chat, ask, edit, and agent
- Visual Studio
- JetBrains IDEs
- Xcode
- Eclipse
- Copilot CLI
- GitHub Copilot cloud agent (the coding agent for delegated tasks)
- github.com and GitHub Mobile
The agent surfaces are where Fable 5 makes the most sense. In simple chat or autocomplete-style use, you’re paying Mythos-class rates for tasks that Sonnet-tier models handle fine; in agent mode on long multi-file tasks, the model’s long-horizon autonomy and lower turn count actually pay for themselves.
How to Enable Claude Fable 5 (Admins)
For Copilot Business and Enterprise, the model is off by default:
- Go to your organization or enterprise Copilot policy settings.
- Find the Claude Fable 5 model policy and set it to enabled.
- Note the data-retention disclosure in the policy description — this is the consent step for the 30-day retention requirement described below.
- Allow time for the rollout; availability may be staged even after the policy is on.
Individual Pro+ and Max users skip this step: when the rollout reaches your account, the model simply appears in the picker.
Billing: Usage-Based, Not a Standard Multiplier
This is the part that surprises teams. According to GitHub’s changelog, Claude Fable 5 is billed at provider list pricing under usage-based billing — meaning consumption maps to Anthropic’s token rates ($10 per million input tokens, $50 per million output tokens) rather than the familiar flat premium-request multipliers used for models like Sonnet or Opus.
Practical implications:
- Agent runs are the cost driver. A long autonomous session in agent mode can consume far more tokens than a chat exchange. Budget per task, not per request.
- Set spending limits. If your org uses usage-based billing budgets, confirm caps before developers start pointing Fable 5 at multi-hour agent tasks.
- Route by difficulty. Keep included or 1×-multiplier models for routine completions and reserve Fable 5 for the migrations, refactors, and debugging marathons where cheaper models stall. Our Fable 5 vs Opus 4.8 comparison covers when the premium is worth it.
Data Retention: The One Claude Model Without ZDR
Every other Claude model in GitHub Copilot — Opus 4.8, Sonnet, Haiku — operates under zero data retention (ZDR). Claude Fable 5 is the exception: prompts and outputs may be retained for up to 30 days so Anthropic can operate the safety classifiers that make a Mythos-class model safe to ship broadly. GitHub states retained data is not used to train Anthropic’s models.
Before enabling the policy, compliance teams should confirm:
- whether company source code and embedded secrets may sit in a 30-day retention window;
- whether client contracts or regulators mandate ZDR for any repositories;
- whether developers need an internal notice when selecting Fable 5 in the picker;
- whether some orgs in the enterprise should keep the policy disabled while others enable it.
If ZDR is a hard requirement, Claude Opus 4.8 in Copilot remains the strongest Claude option without the retention trade-off.
When Fable 5 Is Worth Selecting in Copilot
Reach for Claude Fable 5 in the picker when the task is long, cross-cutting, or failure-prone: large codebase migrations, multi-file refactors with many dependencies, debugging across logs, tests and source, deep pull-request review, or delegated tasks for the Copilot cloud agent. Anthropic’s launch benchmarks show the model’s biggest gains exactly there — 80.3% on SWE-bench Pro versus 69.2% for Opus 4.8.
Skip it for renames, small helpers, stack-trace explanations, and boilerplate tests — cheaper models complete those reliably, and usage-based billing makes habitual Fable use expensive fast.
For a broader look at where Copilot fits against terminal-first agents, see Claude Code vs Codex and Claude vs ChatGPT for coding.