Claude Fable 5.2 in GitHub Copilot needs a retention opt-in

GitHub made Claude Fable 5.2 generally available in GitHub Copilot on November 9, 2026, for Copilot Pro+, Max, Business and Enterprise. In Business and Enterprise organizations, the model stays off until an administrator turns on its policy, and GitHub says turning it on "constitutes acknowledgment" that Anthropic retains prompts and outputs from Fable 5.2 requests. The decision in front of Copilot admins is whether to accept that retention now, ask for a zero-data-retention exemption first, or keep Fable 5.2 off.

GitHub says other Claude models, apart from Fable 5, 5.1 and 5.2, continue to run under zero data retention. So enabling Fable 5.2 changes the data terms for those requests only, and those requests cost twice as much per input and output token as Claude Opus 5.5.

What changed with Claude Fable 5.2 in GitHub Copilot?

Fable 5.2 is rolling out to the Copilot model picker, with its own admin policy that starts disabled for Business and Enterprise. The GitHub changelog lists the model picker in VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, github.com, GitHub Mobile, Copilot CLI, the Copilot app and the Copilot coding agent. Rollout is gradual, so some users won't see it yet.

Anthropic released Claude Fable 5.2 the same day. Anthropic says it generates output up to 30% faster than Fable 5.1 and costs less: $8 per million input tokens and $40 per million output tokens, down from $10 and $50. GitHub says that in early testing Fable 5.2 solved more tasks than Fable 5.1 with fewer steps and tool calls. Both are vendor claims; GitHub's post gives no task counts.

Turning on the policy accepts Anthropic's data retention

Enabling the Fable 5.2 policy is a data-handling decision as well as a model choice. GitHub's changelog says that when Fable 5.2 is used, "Anthropic retains data, including prompts and outputs, to operate safety classifiers that detect harmful use," and that retained data is not used to train Anthropic's models.

Anthropic's own pages fill in the details. Its Covered Models page lists Fable 5.2 as a covered model as of November 9, 2026, with a "30-day minimum retention by default." Prompts and completions are deleted after that unless they are part of a safety investigation or must be kept for legal reasons. Retained data goes through automated safety review. Anthropic's retention practices page adds that its staff cannot read retained conversations by default. Human review happens only through a controlled path, for example after an automated flag, and every access is recorded in a tamper-proof log.

BaristaLabs interpretation: in a coding assistant, the prompt usually carries the code and files the client attaches as context. Treat "prompts and outputs" as able to include any code or file content a Copilot client sends with a Fable 5.2 request, and check that against your customer contracts and internal data rules before enabling it.

Does the Default availability setting turn it on?

No. GitHub's supported models reference says Business and Enterprise plans have a policy that decides whether new, unconfigured GA models are enabled by default. Fable 5, 5.1 and 5.2 are listed as out of scope because they are "not covered by GitHub's data retention agreement." They stay disabled "regardless of your 'Default availability' policy setting."

That matters if your organization delegates new models to an enabled default. For background, see our post on GitHub's global Copilot model policy. Fable 5.2 will not appear through that path. It needs an explicit yes from an administrator.

How much does Claude Fable 5.2 cost in Copilot?

Fable 5.2 is billed at Anthropic's list price under Copilot's usage-based billing. GitHub's models and pricing page converts token costs into AI credits at $0.01 each. Business and Enterprise plans pool each user's included credits at the billing-entity level and bill usage beyond the allowance at these per-token rates (per 1 million tokens):

Scroll sideways to see all 6 columns.

Model in Copilot	Input	Cached input	Cache write	Output	Data retention
Claude Fable 5.2	$8.00	$0.20	$10.00	$40.00	Anthropic retains by default
Claude Fable 5.1	$10.00	$0.25	$12.50	$50.00	Anthropic retains by default
Claude Opus 5.5	$4.00	$0.20	$5.00	$20.00	Zero data retention
Claude Sonnet 5.5	$2.00	$0.10	$2.50	$10.00	Zero data retention

BaristaLabs arithmetic, uncached: a long agent request that reads 200,000 input tokens and writes 20,000 output tokens costs about $2.40 (240 AI credits) on Fable 5.2 and $1.20 (120 credits) on Opus 5.5. Fable 5.2 is cheaper than Fable 5.1 per token, but its input and output rates are still twice those of Opus 5.5.

Anthropic argues the per-task picture can differ. It says Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort on its FrontierCode 2.0 evaluation "for roughly half the cost per task." That is Anthropic's own early testing, and it doesn't tell you how Anthropic's effort levels map to Copilot's reasoning settings. Anthropic also recommends starting most coding work on Opus 5.5 and moving up to Fable when the task needs it. For how Copilot's included allowances work at the plan level, see our breakdown of when GitHub Enterprise's bundled Copilot costs less.

Zero data retention is a request, not a setting

Eligible enterprises can run Fable 5.2, 5.1 and 5 with zero data retention through the end of 2026, and requests go through the GitHub account team. GitHub Support "cannot determine eligibility or enable access," and approval does not switch the model on. An administrator still has to enable the policy afterwards. Once an approved enterprise enables the model, GitHub routes those requests to zero-data-retention endpoints.

The exemption is temporary. GitHub says that after the end of the calendar year, continued use will require Anthropic's Enterprise Frontier Safeguards (EFS). EFS pairs automated safety monitoring with options to keep data, storage and encryption keys under the customer's control. Anthropic describes the ZDR arrangement as "a transition to EFS" and says EFS rolls out in phases beginning in fall 2026. For more on EFS, see our post on Anthropic's Enterprise Frontier Safeguards.

Who should enable Fable 5.2 now

BaristaLabs recommendation:

• Enable it now if your organization already accepts vendor-side retention for AI tools, for example if your teams use Claude on standard terms elsewhere, and you have long, multi-repository tasks where Opus 5.5 falls short.
• Request the ZDR exemption first if your contracts or regulators require zero retention for source code. Ask the account team whether you qualify before anyone enables the policy, and plan for EFS in 2027 rather than treating the exemption as permanent.
• Leave it off if nobody has a task that Opus 5.5 fails on. Fable 5.2 costs twice as much per input and output token and changes your data terms. That trade is easiest to justify on work the cheaper model can't finish.

Teams that already route Fable models through the Claude API can compare this decision with our post on Claude Fable 5 routing policy.

Recommended first step: one team, one written decision

BaristaLabs recommendation: treat the policy switch as a short approval, not a model-picker preference.

1. Write down who approves. The policy toggle is the acknowledgment, so whoever enables it is accepting Anthropic's retention for the organization.
2. Check the data question with whoever owns customer contracts: can prompts that contain your source code be retained by Anthropic for at least 30 days?
3. Ask the GitHub account team about eligibility if the answer is no, and wait for approval before enabling the policy.
4. Enable it for one team with hard, long-horizon work, and review its AI credit use after two weeks against the same work on Opus 5.5.
5. Set a calendar reminder for December if you use the ZDR exemption, since it ends with the calendar year.

For help writing that approval and sizing the credit impact, see AI consulting.

Sources and scope

• Claude Fable 5.2 in GitHub Copilot, GitHub Changelog, November 9, 2026
• Models and pricing for GitHub Copilot, GitHub Docs, retrieved November 10, 2026
• Supported AI models in GitHub Copilot, GitHub Docs, retrieved November 10, 2026
• Introducing Claude Fable 5.2 and Claude Mythos 5.2, Anthropic, November 9, 2026
• Covered Models and Data retention practices for Covered Models, Anthropic Help Center, retrieved November 10, 2026

BaristaLabs did not run Claude Fable 5.2 in Copilot for this post. Prices, retention terms and availability come from GitHub's and Anthropic's pages as of November 10, 2026. GitHub's changelog does not state a retention period; the 30-day figure is from Anthropic's Covered Models page. The cost example is arithmetic from published rates, not a measured bill. Lines marked BaristaLabs interpretation, arithmetic or recommendation are BaristaLabs' views, not GitHub's or Anthropic's.