GitHub Rolls Out Haiku 5.5 in Copilot for Quick Edits and Smaller Tasks

Quick edits and smaller coding assignments have a new model option inside GitHub Copilot. GitHub made Claude Haiku 5.5 generally available on October 7, 2026, targeting fast, high-volume work rather than only long-running jobs. Access is rolling out gradually across five paid plans, so eligible subscribers may not see it immediately.

A small model for smaller assignments

Anthropic introduced the model the same day, positioning it for repeated, cost-sensitive work such as summaries, database queries and classification. It also describes Haiku 5.5 as a subagent—a model handling a smaller assignment within a larger coding project led by another model.

GitHub’s rollout brings that role into Copilot, where it highlights subagents, quick edits and terminal tasks. GitHub says early testing found Haiku 5.5 matched Claude Sonnet 5 on many coding tasks while using significantly fewer tokens and steps. That comparison is with Sonnet 5, not Sonnet 5.5.

Anthropic’s own results show a gap on more demanding work. It reports a 39.2% score for Haiku 5.5 on Terminal-Bench 4.0, a test of complex command-line tasks, against 70.6% for Sonnet 5.5. The company draws a clear boundary around the smaller model’s intended use:

Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding tasks like those measured by Terminal-Bench 4.0.

Anthropic, Claude Haiku 5.5 launch announcement

The rollout reaches editors and agents

Eligible plans are Copilot Pro, Pro+, Max, Business and Enterprise. Users select Haiku 5.5 through the model picker as the rollout reaches their accounts. Supported editors include Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode and Eclipse.

The same option also reaches Copilot CLI, GitHub Copilot’s cloud agent and app, github.com, and GitHub Mobile on iOS and Android. The release therefore spans editor-based work, terminal sessions and cloud-agent assignments rather than being confined to one Copilot client.

The rate card and the access switch

GitHub bills the model at provider list pricing under usage-based billing. Anthropic lists input at $0.10 per million tokens and output at $0.50 for prompts up to 100,000 tokens. Above that threshold, the rates rise to $0.50 and $2.50 respectively. Tokens are the units into which the model divides content.

Those rates are 90% below Haiku 4.5 for shorter prompts and 50% below it for longer ones. Anthropic estimates average running costs fall around 75%, accounting for request sizes and a changed tokenizer that uses slightly more tokens per task. Lower token prices are not the same as identical savings on finished work.

For organizational users, availability also depends on policy. Business and Enterprise administrators can manage Haiku 5.5 in Copilot settings. New models are enabled automatically unless an administrator has turned off global default enablement or explicitly disabled this model.