Haiku 5.5 arrives in GitHub Copilot: eligible plans and why it may not appear yet

The new fact is Haiku 5.5 selection inside Copilot

GitHub announced Claude Haiku 5.5 general availability in Copilot on October 7, 2026. It adds a model aimed at fast, frequent work; GitHub's examples include subagents, small edits, and terminal tasks.

Supported plans are Pro, Pro+, Max, Business, and Enterprise, not Free. Legacy annual Pro/Pro+ accounts retaining request-based billing are an exception: GitHub says those plans do not receive new models and features. A similarly named Claude plan or API permission cannot replace a Copilot subscription.

This differs from Anthropic's model and API-price launch. Availability of a model does not mean every tool integrates it that day. GitHub's announcement establishes Copilot entry points and organization control.

Check the model picker in the client you use

Officially listed clients include Visual Studio Code, Visual Studio, Copilot CLI, Copilot cloud agent, Copilot app, and github.com. Mobile includes iOS and Android; other IDEs include JetBrains, Xcode, and Eclipse.

Open the client's model picker and look for Claude Haiku 5.5. The announcement gives no minimum version for each client or simultaneous availability promise. Rollout is explicitly gradual.

If absent, check the signed-in account and actual Copilot plan, then organization policy. If both qualify, rollout may still be pending. Model absence alone does not justify rebuilding the entire development environment.

Organization model policy still controls access

Business and Enterprise administrators manage Haiku 5.5 through Copilot model policies. New models enable automatically under the default model rule unless the administrator disables the global default or the model explicitly.

Fictional example: a developer can use the model personally but their organization disables it. Availability personally and absence on an organizational seat can reflect policy, not a network or client failure.

Administrators should review the global setting and the individual policy against data and budget requirements. Model policy differs from the new-feature default policy taking effect October 22; this access decision need not wait until then.

How is the inexpensive model billed inside Copilot?

GitHub says usage-based Haiku 5.5 follows the provider's published pricing. AI Credits are GitHub's billing unit, with one credit equal to $0.01. Included personal allowances vary by plan; Business and Enterprise pool per-seat allowances at the billing entity.

As rechecked October 8, GitHub lists Haiku 5.5 prices. For input up to 100K tokens, per-million-token input, cache-read, cache-write, and output rates are $0.10, $0.01, $0.125, and $0.50 respectively. Caching reuses or stores context.

For input above 100K tokens, the corresponding rates are $0.50, $0.05, $0.625, and $2.50 per million. The tier depends on input length, not output or combined input and output. These are GitHub's current usage rates, not an inferred task total from Anthropic pricing.

Before using it, inspect actual personal or organizational billing, included allowance, and extra-usage settings. Lower model rates can reduce total cost only when evaluated with input, output, tool steps, and retries.

Can legacy annual request-based plans obtain this model?

After GitHub moved to usage-based billing June 1, some existing annual Pro and Pro+ subscriptions retained request-based billing. These accounts continue deducting requests with applicable multipliers; a new launch does not change their billing system automatically.

More importantly, legacy annual documentation says they do not receive new models and features, and its multiplier table does not list Haiku 5.5. If request-based billing remains, check plan-change options instead of attributing absence solely to rollout or applying Haiku 4.5's multiplier.

GitHub also says users who subscribed to Copilot through GitHub Mobile on iOS or Android cannot purchase additional AI Credits. If extra credits are central to your plan, check subscription channel and available account options first.

Start with a small, clearly scoped task

Explaining an error, editing an existing function, or handling a bounded subtask are hypothetical starting points, not tests performed here. Keep the prior workflow available and compare quality and actual consumption before switching all tasks.

After selecting the model, verify the task's displayed model and account usage records. Broad refactors, long planning, or repeated rework still warrant comparison with other models. GitHub's early testing description does not guarantee equal outcomes for every project.

The useful sequence is plan and billing verification, model-picker inspection, organization-policy review, then observed consumption. Client rollout timing depends on later official updates and the account; use current GitHub pricing and actual input length.