September 19, 2026
Auto mode in Claude Code is cheaper for API: v2.1.278 moves checks to the server
On September 19, 2026, Claude Code v2.1.278 enabled the server-side classifier by default for Claude API, Enterprise, Claude Platform on AWS, Bedrock, Google Cloud Agent Platform, Microsoft Foundry and LLM gateways. This review adds no classifier overhead charge, and the CLI warns about costs when it falls back to a billable alternative.
On August 7, 2026, the company removed classifier overhead for Pro, Max and Team. For Enterprise, Claude API, Bedrock, Vertex and Foundry, the server-side setup remained opt-in at the time.
Now the review is universal. Claude Code requests server-side review as part of regular model requests. If server-side review is unavailable because of a gateway, region, platform or credentials, the CLI sends its own requests to the classifier and warns about costs where they are billed. `/status` now includes an Auto mode server line showing the review method for the current session.
A signal from Anthropic. Auto mode users in Teams and Enterprise ship roughly 25% more PRs. Adobe, Nuro, Gusto and Garner Health use the mode in production. Garner Health enabled Auto mode through managed settings for 550 employees, while about 10% of Gusto sessions since mid-May included a classifier denial.
Setup. In the CLI, the mode is enabled with `claude --permission-mode auto`. For Bedrock, Google Cloud Agent Platform and Microsoft Foundry, the default is set through `permissions.defaultMode: "auto"` in user or managed settings. On macOS and Linux, Claude Code is installed with `curl -fsSL https://claude.ai/install.sh | bash` or via `brew install --cask claude-code`, then launched with `claude` in the project directory. For Windows, use `irm https://claude.ai/install.ps1 | iex` or the `winget install Anthropic.ClaudeCode` package.
On a direct connection to the Anthropic API, the `CLAUDE_CODE_AUTO_MODE_SERVER=0` variable is not read.
