Claude Haiku 5.5 in GitHub Copilot: Availability, Pricing and Admin Controls

What GitHub announced​

GitHub says Claude Haiku 5.5 is generally available in Copilot. It is built for fast, high-volume work such as subagents, quick edits, and terminal tasks. GitHub also says that in early testing Haiku 5.5 matched Claude Sonnet 5 on many coding tasks while using significantly fewer tokens and steps.

Treat that last claim as GitHub's own characterization. The announcement gives no benchmark names, task lists, token counts or methodology. It is a reason to run your own trial, not proof of parity. Your results will depend on your codebase and prompts.

Where and for whom​

The model is available to Copilot Pro, Pro+, Max, Business, and Enterprise users. GitHub lists these places where it can be picked from the model picker:

• Visual Studio Code
• Visual Studio
• Copilot CLI
• GitHub Copilot cloud agent
• GitHub Copilot app
• github.com
• GitHub Mobile on iOS and Android
• JetBrains IDEs
• Xcode
• Eclipse

The rollout is gradual, so a missing model right after the announcement does not by itself mean something is wrong with your account. Check back later. If it still doesn't appear, look at your plan, your client and, on organizational plans, your admin's model policy.

Windows developers should also keep their editors current. GitHub's supported-models documentation keeps a table of minimum IDE versions for recent models. It lists, for example, VS Code v1.124 and Visual Studio 17.14.6 for Claude Sonnet 5. The snippet I reviewed had no entry for Haiku 5.5, so I can't give you a minimum version for it. If the picker stays empty on an old build, updating is a sensible first step.

Pricing: the part worth reading closely​

GitHub says the model is billed at provider list pricing under usage-based billing. GitHub's models and pricing documentation lists Haiku 5.5 as GA and Lightweight. Prices are per million tokens:

GitHub's table shows how these rates compare with other Claude models in Copilot:

• Claude Haiku 4.5: $1.00 input and $5.00 output, with no long-context tier.
• Claude Sonnet 5: $2.00 input and $10.00 output.
• Claude Sonnet 5.5: $2.00 input and $10.00 output.

On default-tier list rates, Haiku 5.5 is therefore about a tenth of Haiku 4.5's price, and a twentieth of Sonnet 5's. That gap is large. Combined with GitHub's claim of fewer tokens per task, it would make the model attractive for high-volume agent work if the quality holds. Both halves of that equation are still unproven for your workload.

A few details are easy to get wrong:

• Cost isn't per request. It depends on how many input, output, cached and cache-write tokens an interaction uses. Anthropic models carry a cache-write charge on top of cached-input pricing.
• Long context costs more. Once a request's input passes 100K tokens, the higher tier applies.
• Credits are converted. GitHub converts token costs into AI Credits at 1 credit = $0.01.
• Plans include allowances. Individual plans have included credit allowances that vary by plan. Business and Enterprise plans have per-user allowances pooled at the billing-entity level. Charges for extra usage only start once those allowances run out.
• Mobile subscribers can't top up. Buying additional AI credits isn't available if you subscribe, or ever subscribed, to Copilot through GitHub Mobile on iOS or Android.

Code completions and next edit suggestions aren't billed in AI Credits, so the model's price doesn't touch those features.

Admin controls for Business and Enterprise​

Business and Enterprise administrators manage access through the model policy in Copilot settings. Under default model enablement, new models switch on automatically unless an administrator has turned off the global default or explicitly disabled that model.

GitHub's policy documentation adds that this default applies to unconfigured GA models. Enterprise policies are set first, and organizations may be allowed to decide for themselves. Where a user holds licenses from several organizations, the least restrictive policy usually applies, with exceptions.

Teams with strict governance should consider:

1. Deciding whether the default-on setting suits your cost and compliance posture before models like this arrive.
2. Explicitly disabling Haiku 5.5 if you want to evaluate it first.
3. Reviewing who can edit policies and watching the audit log for changes. GitHub's documentation recommends both to prevent policy drift.

Because the model is cheap per token, defaulting it on carries little cost risk. Teams will still care about which models touch their code, and the cheaper rate might tempt people to run more agent work, which increases total token use.

A practical way to evaluate it​

This is my suggestion, not GitHub's guidance:

1. Confirm the model appears in your picker and your plan and policy allow it.
2. Pick a few representative tasks: small edits, terminal help, a subagent job.
3. Run them on Haiku 5.5 and on the model you use now, and compare output quality and how many retries each needs.
4. Check token usage and AI Credit consumption in your usage reports, since the lower rate only helps if results don't need rework.

GitHub's announcement establishes that the model is available, who can use it, where, and at what list price. It does not establish how much faster it is, which tasks suit it best, or how much a team would actually save.