GitHub announced the general availability of Claude Haiku 5.5 in GitHub Copilot on October 7, 2026. Anthropic’s lightweight model is aimed at fast, high-volume work, including subagents, quick edits, and terminal tasks.
For developers, the change adds another model-selection option for latency-sensitive or repetitive workflows. For organizations, it also introduces the usual questions around model policies, AI-credit consumption, client support, and gradual rollout. GitHub says early testing found Haiku 5.5 matched Claude Sonnet 5 on many coding tasks while using significantly fewer tokens and steps, but the supplied announcement does not include independent benchmark methodology or results.
What changed in GitHub Copilot
Claude Haiku 5.5 is now listed as generally available in GitHub Copilot. The announcement includes Copilot Pro, Pro+, Max, Business, and Enterprise users within the availability scope.
GitHub positions the model for tasks where responsiveness and throughput matter more than selecting the largest available model. Examples in the announcement include subagents, quick edits, and terminal work. That makes Haiku 5.5 a candidate for repetitive changes, short interactions, and workflows that may generate many model requests.
Users select the model from the Copilot model picker where it is exposed. Availability is not determined by plan alone: GitHub says it also depends on the client or surface where Copilot is being used. The supported-models documentation lists Haiku 5.5 in its client table for GitHub.com, Copilot CLI, Visual Studio Code, Visual Studio, Eclipse, Xcode, and JetBrains IDEs.
The announcement also names the GitHub Copilot app and GitHub Mobile on iOS and Android among the selection surfaces, and says the rollout is gradual. The broader announcement and the maintained per-client table serve different purposes: the announcement describes the overall rollout scope, while the reference documentation provides more specific client-level availability. Neither should be read as proof that every platform exposes identical features at the same time.
Billing depends on tokens and context
Haiku 5.5 uses GitHub’s token-based AI-credit billing model. GitHub documents one AI credit as equal to $0.01 USD, with charges based on input tokens, output tokens, and cached tokens.
For the Default pricing tier, GitHub lists the following Haiku 5.5 rates per million tokens:
- $0.10 for input tokens
- $0.01 for cached input tokens
- $0.125 for cache writes
- $0.50 for output tokens
The Long context tier is listed at $0.50 for input, $0.05 for cached input, $0.625 for cache writes, and $2.50 for output per million tokens. GitHub says Haiku 5.5 uses the Default tier for inputs of 100,000 tokens or fewer and the Long context tier above 100,000 input tokens. Anthropic models also have a cache-write cost in addition to cached-input pricing.
These rates do not establish a fixed cost per prompt. Actual usage depends on the number of tokens processed, caching, context length, the Copilot feature being used, plan allowances, and any additional-usage settings. Agentic workflows may consume more credits than short interactions because they can involve multiple model calls.
GitHub’s billing documentation currently lists monthly individual allowances of 1,500 AI credits for Copilot Pro, 7,000 for Pro+, and 20,000 for Max. For organizations and enterprises, the documented allowances are 1,900 credits per Copilot Business user and 3,900 per Copilot Enterprise user per month. Business and Enterprise credits are pooled at the billing-entity level.
Copilot Chat, Copilot CLI, Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents consume AI credits. Code completions and next-edit suggestions are not billed in AI credits and remain unlimited for paid plans, according to GitHub’s individual and organization billing documentation.
What administrators need to review
Business and Enterprise administrators can control Copilot model access through enterprise or organization policies. GitHub documents a models policy that can determine whether generally available models default to enabled or disabled when they have not been explicitly configured.
Policies can apply across authenticated Copilot surfaces, including IDEs, GitHub’s website, and Copilot CLI. However, not every policy applies to every surface, and the GitHub Copilot app and Copilot CLI have separate client policies. Administrators should therefore check the relevant client settings rather than assuming that one organization-level choice produces identical behavior everywhere.
Teams should also account for pooled usage. A model that is inexpensive per token can still affect the organization’s budget when used by agents or other high-volume features. Reviewing AI-credit consumption, additional-usage rules, and model defaults is particularly important before enabling the model broadly.
Context and client limitations remain important
GitHub’s supported-models reference associates Claude Haiku 5.5 with extended-capability documentation that includes a 1 million-token context-window option. The documentation says that option is available only in Visual Studio Code and Copilot CLI. However, the supplied documentation does not provide a complete standalone Haiku 5.5 context-window specification or fully clarify every extended capability by client.
The documented 100,000-token billing threshold should not be confused with a complete statement of the model’s context window. It describes when GitHub applies the Long pricing tier, not necessarily the maximum context available on every platform.
The same reference does not provide a Haiku-specific minimum IDE or extension version; the corresponding entries are marked TBD. Keeping Copilot clients, IDEs, and extensions current is still a sensible operational step, but teams should not infer a minimum supported version from the available documentation.
What developers should do
- Check the model picker. Confirm whether Claude Haiku 5.5 is visible in the Copilot client used for the target workflow. General availability does not mean the gradual rollout has completed for every account or surface.
- Match the model to the task. Start with quick edits, terminal tasks, repetitive coding assistance, or other high-volume workflows where a lightweight model may be appropriate. Do not assume that GitHub’s early Sonnet comparison guarantees equivalent results for every codebase or task.
- Track AI-credit use. Review usage for Chat, CLI, cloud agent, Spaces, Spark, and third-party coding agents separately from unlimited paid-plan completions and next-edit suggestions.
- Review organizational controls. Business and Enterprise administrators should inspect model policies, default enablement, client-specific policies, budgets, and additional-usage settings before making Haiku 5.5 broadly available.
- Test across required clients. Availability and feature behavior can vary by plan and client. Validate the workflows that matter instead of assuming that GitHub.com, an IDE, CLI, mobile, and agent surfaces behave identically.
Availability and documentation caveats
Claude Haiku 5.5 is generally available, but GitHub describes its rollout as gradual. Users may therefore need to wait before the model appears. Access remains subject to the Copilot plan, client, and applicable organization or enterprise policies.
GitHub’s default AI-model prompts and output completions pass through content filters for harmful, offensive, or off-topic content, as well as public-code matching when enabled. The supplied documentation does not make Claude Haiku 5.5-specific security, data-retention, compliance, or privacy claims.
Measured latency improvements, model-specific quotas, exact rollout completion dates, and independent validation of the Sonnet comparison are not documented in the supplied sources. Those limitations matter when teams estimate performance or cost: Haiku 5.5 is positioned as a fast, economical option, but the results should be validated against the organization’s own workloads.
