Image: GitHub
Why it mattersA small, fast model in the same picker as Sonnet lets a team send the many tiny edits and terminal calls to the cheaper model without leaving the editor.
Picking a smaller model for a tiny edit used to mean leaving the editor. GitHub Copilot now ships Claude Haiku 5.5 inside the model picker on every paid tier.
The GitHub Changelog says Haiku 5.5, Anthropic's newest small model, is generally available in Copilot for Pro, Pro+, Max, Business and Enterprise users. GitHub says it is built for "fast, high-volume work like subagents, quick edits, and terminal tasks". In early testing by GitHub, Haiku 5.5 "matched Claude Sonnet 5 on many coding tasks while using significantly fewer tokens and steps".
Where it shows up
The model is selectable in Visual Studio Code, the Copilot CLI, github.com, the Copilot app, GitHub Mobile on iOS and Android, JetBrains IDEs and Xcode. The rollout is gradual, so a user who does not see it yet is told to check back soon.
The entry is billed under usage-based billing at provider list pricing. GitHub links its models and pricing page for the figures and does not restate them in the changelog.
Admin controls on business plans
Enterprise and Business plan administrators can allow or block Haiku 5.5 through the model policy in Copilot settings. New models are automatically enabled unless an administrator has turned off the global default or explicitly disables this model, so a plan with default enablement switched on will pick up Haiku 5.5 the moment a user's rollout flips. A plan with it off has to opt the model in by hand.
The GitHub Changelog does not state a Haiku 5.5 price in that post, name the comparison harness behind the Sonnet 5 claim, or publish token count figures. GitHub's own list-pricing page is the only place these facts sit, and a price change there would not update the changelog.
The claim that Haiku 5.5 matches Sonnet 5 on "many coding tasks" is GitHub's own, measured in GitHub's "early testing", and the changelog does not say which tasks, which codebases, or how many runs. Copilot's price for using the model inside the editor sits on the Copilot models and pricing page, quoted in premium requests, which is a different unit from the raw tokens-per-million figure Anthropic lists for the API. A team wanting a per-request estimate has to read both.
A team already paying for Copilot can now split a job between two models in one picker: Sonnet or Opus for the design, Haiku 5.5 for the ten follow-up edits and the terminal probes. The practical opening is subagents, which the model is explicitly aimed at: a Sonnet-driven agent that fans out ten small lookups through Haiku pays for the small model ten times and the large one once, instead of the other way round. Whether that saves anything depends on how much of the day is small work, which is the number only a Copilot usage report will show. Enterprise administrators can see the breakdown in the agent activity usage metrics page.
Source
- Claude Haiku 5.5 in GitHub Copilot, GitHub Changelog
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.