Claude Fable 5.1 in GitHub Copilot: Plans, Default Off, Data Retention

GitHub’s changelog on 2026-09-01 announced that Claude Fable 5.1 is generally available in GitHub Copilot.

Most coverage frames this as one more model in the picker. The line that actually requires a decision is a different one: GitHub states that Fable 5.1 requires data retention by default, and that this differs from the other Claude models on the platform. For regulated teams, or anyone with source-code confidentiality obligations to customers, that sentence outranks the question of how good the model is.

1. Who can select it today

Eligible plans per the changelog: Copilot Pro+, Max, Business and Enterprise.

Copilot Pro (US$10/month) is not on the list, and neither is Free. Against the current plan page:

Surface coverage is broad: VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, the Copilot CLI, the GitHub Copilot app, GitHub.com, and GitHub Mobile on iOS and Android. GitHub notes a gradual rollout, so an eligible plan without the model in the picker yet is normal.

2. For orgs, it ships switched off

The second place teams get stuck: GitHub states that Business and Enterprise administrators must manually enable the Claude Fable 5.1 policy in Copilot settings, because it is disabled by default.

So if enterprise users cannot find it in the VS Code model list, the usual cause is not the rollout — it is that nobody has flipped the policy.

3. The actual news: data retention

GitHub’s wording is that Fable 5.1 requires data retention by default, in order to run Anthropic’s safety classifiers — unlike the other Claude models available in Copilot.

The announcement also gives an exception path:

• Eligible enterprise customers can get zero data retention through year-end, while Anthropic builds out Enterprise Frontier Safeguards (EFS).
• To find out whether you qualify, you must contact your GitHub account team.
• And explicitly: approval does not automatically enable access — an admin still has to turn the policy on.

What that means in practice:

1. If your company has an internal rule or a customer contract saying source code must not be retained by third parties, do not enable the policy before the zero-retention question is settled.
2. Zero data retention is a time-boxed arrangement (“through year-end” in GitHub’s words), not a permanent state. What happens after that is not addressed in this announcement.
3. Enablement is two-stage. Eligibility approval and the admin switch are separate events, and the first does not trigger the second.

4. Billing: it does not use the old multiplier table

Many teams still ask “how many premium requests is this model worth”. As of 2026-06-01, GitHub moved to usage-based billing measured in AI credits, and the model multiplier table now applies only to Copilot Pro and Pro+ users who stayed on the legacy annual request-based plan after that date.

Fable 5.1 sits on the new model and is billed at provider list pricing, not through a multiplier. So:

• On a legacy annual plan, you cannot select Fable 5.1 anyway, and the multiplier question is moot.
• On usage-based billing, credits are consumed as you go, and the thing to watch is what happens when the included allowance runs out.

5. One sentence each, for three kinds of reader

• Individual Pro users (US$10): this release is not for you. Fable 5.1 means moving to Pro+ at US$39. Decide whether one model is worth tripling the subscription.
• Enterprise admins: settle the data-retention question first, then decide about enabling. Doing it in the other order trades your company’s source code for a model option.
• Existing Claude subscribers: Fable 5.1 inside Copilot and a direct Claude subscription are two separate bills under two separate policies.

6. What we could not verify

• The effective unit cost of Fable 5.1 inside Copilot: GitHub points to provider list pricing without a Copilot-specific per-token table. Unverified.
• What happens after the “through year-end” zero-retention window: not addressed. Unverified.
• The eligibility bar for “eligible enterprise customers”: you are told to contact an account team; no public criteria. Unverified.
• When the gradual rollout completes: not stated. Unverified.
• Quality comparisons against other models: GitHub published no citable benchmark, and we do not invent scores.

Official links

• GitHub Changelog (2026-09-01): https://github.blog/changelog/2026-09-01-claude-fable-5-1-generally-available-in-github-copilot
• Copilot plans and allowances (prices and credits above come from here): https://docs.github.com/en/copilot/get-started/plans
• Supported models in Copilot: https://docs.github.com/en/copilot/reference/ai-models/supported-models
• Legacy request-based billing concepts: https://docs.github.com/en/copilot/concepts/billing/copilot-requests

Related

• GitHub Copilot tool page (plans and quota verification log)
• Claude tool page
• What Copilot free and paid actually differ on
• Keep .env away from Copilot: content exclusion setup