Auto Mode
A model-dispatching strategy in AI programming tools whereby the system automatically selects an appropriate model based on task complexity: lightweight models handle simple tasks while high-performance models are invoked for complex ones, achieving a balance between output quality and cost.
Timeline (last 90 days)
Auto mode previously retried indefinitely when a safety check refused to review an operation; after the fix, it now refuses once directly and explains that retrying is unhelpful.
In version v2.1.278, the Auto mode classifier defaults to running server-side, and this overhead is no longer billed to users.
The introduction of the server-side classifier means that Anthropic bears the classification compute costs on its own infrastructure, rather than passing them through as fees to downstream cloud platform users.
Users running on Bedrock, Vertex, Foundry, and gateways can disable the server-side classifier and revert to the original behavior by setting the environment variable CLAUDE_CODE_AUTO_MODE_SERVER=0.
The server-side classifier default switch applies to Claude API and enterprise users, as well as deployments running on Amazon Bedrock, Google Vertex, Foundry, and various gateways.
Under the previous mechanism, the computational overhead of Claude Code's Auto mode classifier was counted toward the user's bill.
Users can enter Cursor's settings to check the current model configuration, disable Auto mode, and manually specify a model to ensure output consistency.
Under the new billing logic, there is no cost difference between a user manually selecting a specific model and Auto mode automatically invoking that same model.
Auto mode may autonomously use an extremely expensive model and exhaust a user's entire token quota in a single task.
When Cursor runs in Auto mode, it does not clearly indicate which model is being used, and users can only learn which models were invoked after the fact through the usage history in the dashboard.
4 more timeline events
All Facts (12)
Auto mode means the tool automatically selects the appropriate model to handle tasks — lightweight models for simple tasks and more powerful models for complex tasks — serving as a cost optimization strategy.
70%UnverifiedCursor Auto's billing method will change from a uniform flat rate to pricing based on the actual model each request is routed to.
60%UnverifiedAuto mode previously retried indefinitely when a safety check refused to review an operation; after the fix, it now refuses once directly and explains that retrying is unhelpful.
50%UnverifiedIn version v2.1.278, the Auto mode classifier defaults to running server-side, and this overhead is no longer billed to users.
50%UnverifiedThe server-side classifier default switch applies to Claude API and enterprise users, as well as deployments running on Amazon Bedrock, Google Vertex, Foundry, and various gateways.
50%UnverifiedUsers running on Bedrock, Vertex, Foundry, and gateways can disable the server-side classifier and revert to the original behavior by setting the environment variable CLAUDE_CODE_AUTO_MODE_SERVER=0.
50%UnverifiedThe introduction of the server-side classifier means that Anthropic bears the classification compute costs on its own infrastructure, rather than passing them through as fees to downstream cloud platform users.
50%UnverifiedUnder the previous mechanism, the computational overhead of Claude Code's Auto mode classifier was counted toward the user's bill.
50%UnverifiedUsers can enter Cursor's settings to check the current model configuration, disable Auto mode, and manually specify a model to ensure output consistency.
50%UnverifiedWhen Cursor runs in Auto mode, it does not clearly indicate which model is being used, and users can only learn which models were invoked after the fact through the usage history in the dashboard.
50%UnverifiedThe prerequisite for enabling Auto mode in the build phase is establishing a comprehensive permission policy, including locking down the toolset, restricting web resources, and managing packages.
50%UnverifiedIt is recommended to use the Auto automatic mode in everyday use, which automatically selects a model based on the task and prioritizes consuming the quota of its own models, only temporarily switching to a more powerful model when a particular step repeatedly fails.
50%