Last Updated: 19 May 2026
GitHub Copilot cloud agent is getting more flexible. Users can now choose smaller AI models when they want the agent to handle simple coding work, such as small fixes, updates, or basic changes. GitHub says Anthropic's Claude Haiku 4.5 and OpenAI's GPT 5.4 Mini are now available for the cloud agent, each with a 0.33x multiplier.
That small number is the key. A lower multiplier means these models can be more cost friendly for routine work. For teams that use Copilot often, this gives them a cleaner way to save stronger models for harder jobs and use lighter models for smaller ones.
GitHub Copilot cloud agent lets users give coding tasks to an AI agent that works in the cloud. Instead of only chatting with Copilot, users can assign work and let the agent return changes. GitHub’s docs say model selection is available when starting a task in supported places.
The new update adds Claude Haiku 4.5 and GPT 5.4 Mini to the list of supported models. GitHub describes these as faster and more cost efficient options for simple tasks. They sit beside larger models that are still better suited for harder work.
The current supported options for Copilot cloud agent include Auto, Claude Sonnet 4.5, Claude Opus 4.7, Claude Haiku 4.5, GPT 5.2 Codex, and GPT 5.4 Mini. Auto lets Copilot choose a model for the user when manual selection is not needed or not available.
Not every coding task needs the most powerful AI model. A typo fix, a small file update, a basic refactor, or a short test change may only need a faster model. GitHub is now giving users a way to match the model to the task size.
Claude Haiku 4.5 and GPT 5.4 Mini both have a 0.33x multiplier for paid plans. Basically, they count as less usage than many larger models. GitHub also notes that model multipliers can change, so teams should still check the current billing page when planning usage.
This is useful for developers who send many small tasks to Copilot cloud agent. Instead of spending more model usage on each simple request, they can choose a smaller model and save the heavier options for more complex work.
Smaller tasks now have a cleaner path: pick a lighter model, run the agent, review the result.
GitHub says model selection works in supported entry points. These include assigning an issue to Copilot on GitHub.com, mentioning @copilot in a pull request comment on GitHub.com, starting a task from the agents tab, using the agents panel, GitHub Mobile, or the Raycast launcher.
When the model picker is not available, Copilot uses Auto. That means users do not always need to pick a model by hand. GitHub says Auto model selection chooses based on availability and can help reduce rate limiting.
For many users, Auto will be enough. For teams with cost targets or clear task types, manual model choice gives more control. A small bug fix can go to Claude Haiku 4.5 or GPT 5.4 Mini. A harder task can go to a more capable model.
GitHub Copilot cloud agent is available with GitHub Copilot Pro, GitHub Copilot Pro Plus, GitHub Copilot Business, and GitHub Copilot Enterprise. GitHub says it works in repositories stored on GitHub, unless the repository is owned by managed user accounts or the feature has been disabled.
That means this update is not only for large companies. Individual paid users can also access the cloud agent if they are on a supported plan. For teams, the bigger value is control across repeated tasks.
How to decide wich model you shoul use:
This gives developers a more practical workflow without asking them to learn a complex setup.
The update points to a more careful way of using AI coding agents. Developers can stop treating every task as the same. A small issue can use a smaller model. A deeper code change can use a stronger one.
For teams, the billing angle is important. Copilot is moving from request based billing to usage based billing starting June 1, 2026. That makes model choice more important because different models can carry different multipliers.
The safest habit is still review first. Copilot cloud agent can help produce changes, but developers should inspect the result, run tests, and check whether the change fits the project. A cheaper model is useful only when the output is good enough for the job.
GitHub Copilot cloud agent now gives users more control over how AI handles coding tasks. Claude Haiku 4.5 and GPT 5.4 Mini bring faster, lower multiplier options for simple work, which can help developers avoid using larger models when they are not needed.
The practical value is clear. Pick Auto when you want Copilot to decide. Pick a smaller model for small tasks. Save larger models for work that needs more reasoning. The update makes Copilot cloud agent easier to fit into daily development without adding much complexity.
GitHub Copilot cloud agent is an AI tool that can work on coding tasks in the cloud. Instead of only giving suggestions in a chat or editor, it can take a task and produce code changes for review. Developers can use it for issues, pull request comments, and supported agent entry points. The main idea is to let Copilot handle a defined task, then let the developer inspect the output before using it in a project.
GitHub added two model options for Copilot cloud agent: Claude Haiku 4.5 and GPT 5.4 Mini. These models are meant for simpler tasks where a larger model may not be needed. Both have a 0.33x multiplier for paid plans, according to GitHub’s model page. This gives users a way to use faster and lower cost options for routine coding work, while keeping stronger models available for more complex tasks.
Use Claude Haiku 4.5 or GPT 5.4 Mini when the task is small and easy to check. Good examples include simple bug fixes, basic file changes, small test updates, short documentation edits, and minor code cleanup. These models are not meant to replace larger models for every job. They are better used when the task is clear, limited, and low risk. The final result should still be reviewed before merging.
If you do not choose a model, or if the model picker is not available, GitHub Copilot cloud agent can use Auto. Auto model selection chooses a model for the task based on availability and other system factors. This is useful for users who do not want to think about model choice each time. Manual selection is better when you already know the task is small, complex, or tied to cost planning.
GitHub says Copilot cloud agent is available with Copilot Pro, Copilot Pro Plus, Copilot Business, and Copilot Enterprise. It can be used in repositories stored on GitHub, unless the repository is owned by managed user accounts or the feature has been disabled. Access can also depend on plan settings, organization policies, and model availability. Teams should check their Copilot settings before building workflows around cloud agent tasks.
