Claude Haiku 5.5 Lands in GitHub Copilot
Anthropic's lightweight model reached ten Copilot surfaces on October 7, 2026, at list prices of $0.10 and $0.50 per million tokens.
In short
Claude Haiku 5.5 became generally available in GitHub Copilot on October 7, 2026, across ten client surfaces and for the Pro, Pro+, Max, Business, and Enterprise plans.
At a glance
- Live since October 7, 2026 across ten Copilot surfaces, including VS Code, Copilot CLI, JetBrains IDEs, Xcode, and Eclipse.
- Enabled for Copilot Pro, Pro+, Max, Business, and Enterprise; billed at provider list pricing under usage-based billing.
- List price up to 100,000 tokens: $0.10 and $0.50 per million tokens; above that, $0.50 and $2.50.
- Predecessor Haiku 4.5 cost $1 and $5 per million tokens; Product Hunt cites 90% lower pricing up to 100,000 tokens.
- Simon Willison: the new tokenizer turns the same prompts into roughly 1.25x as many tokens as Haiku 4.5.
Claude Haiku 5.5 became generally available in GitHub Copilot on October 7, 2026, across ten client surfaces and for the Pro, Pro+, Max, Business, and Enterprise plans. GitHub aims it at volume rather than depth: quick edits, terminal work, and fleets of subagents.
Where you can pick it
The changelog lists Visual Studio Code, Visual Studio, Copilot CLI, the Copilot cloud agent, the Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains IDEs, Xcode, and Eclipse. GitHub describes the rollout as gradual, so the model picker will not show Haiku 5.5 everywhere on day one. Usage lands on provider list pricing through usage-based billing.
What the changelog does not state is a premium request multiplier for Haiku 5.5. Neither of the other two sources read for this piece supplies that number either, so it is an open figure rather than a zero.
The pricing case
Up to 100,000 tokens, Haiku 5.5 lists at $0.10 per million input tokens and $0.50 per million output tokens. Past that threshold the rate moves to $0.50 and $2.50. Haiku 4.5 charged $1 and $5 per million tokens, and the Product Hunt listing frames the change as 90% lower pricing for prompts up to 100,000 tokens. The model also ships through the Claude platform, AWS, Google Cloud, and Microsoft Azure.
Simon Willison flags a line item that appears on no price sheet: the new tokenizer counts roughly 1.25x as many tokens for identical prompts as Haiku 4.5 did. Measured per task instead of per token, the cut shrinks.
Effort as a dial
Haiku 5.5 exposes an adjustable effort setting. In Willison's SVG drawing test, the 'low' setting finished in 7 seconds for 0.0936 cents, while 'max' took 5 minutes 9 seconds and cost 3.3826 cents. Anyone fanning out a dozen subagents is therefore making a budget decision per call, not once per project.
What the performance claim covers
GitHub says that in early testing the model kept pace with Claude Sonnet 5 across a good deal of coding work while spending markedly fewer tokens and fewer steps. No benchmark, task count, or methodology accompanies that sentence. It stands as a vendor statement, not a result anyone outside GitHub can check.
Who can switch it off
Business and Enterprise organizations govern access through Copilot's model policy. New models turn on automatically as long as the global default is left in place. Keeping Haiku 5.5 away from a team is an opt-out an administrator has to perform, not an approval they withhold.
FAQ
Which GitHub Copilot clients support Claude Haiku 5.5?
Ten surfaces: Visual Studio Code, Visual Studio, Copilot CLI, the Copilot cloud agent, the Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains IDEs, Xcode, and Eclipse, with a gradual rollout.
How much does Claude Haiku 5.5 cost per million tokens?
Up to 100,000 tokens it is $0.10 input and $0.50 output per million tokens, and $0.50 and $2.50 above that. Haiku 4.5 was $1 and $5. Simon Willison notes the new tokenizer counts about 1.25x as many tokens.
How do Copilot admins block Claude Haiku 5.5 for their organization?
Through the model policy on Business and Enterprise plans. New models are enabled by default while the global default stays on, so an administrator has to set the restriction explicitly.
