Opus 4.6 Becomes the Default on Bedrock, Vertex, and Microsoft Foundry

Claude Code

Claude Code 2.1.73 changes the default Opus model on Amazon Bedrock, Google Vertex AI, and Microsoft Foundry from Opus 4.1 to Opus 4.6. The update automatically applies to all users on these platforms who have not pinned a specific model version, bringing cloud-hosted deployments in line with the flagship capabilities of Opus 4.6 including its expanded reasoning and 1-million-token context window. Anthropic simultaneously deprecated the /output-style command in favor of /config, fixing output style at session start to enable better prompt cache reuse.


Cloud Provider Defaults Align with Opus 4.6

With the release of Claude Code 2.1.73, the default Opus model for Amazon Bedrock, Google Vertex AI, and Microsoft Foundry deployments advances from Opus 4.1 to Opus 4.6. The change takes effect automatically for any organization that has not explicitly pinned a model version — bringing cloud-hosted Claude Code environments up to the same capabilities available to direct Anthropic API users.

What Changes for Cloud Platform Users

Organizations running Claude Code through AWS, GCP, or Azure-based deployments will now see Opus 4.6 invoked by default when the Opus family is selected. Opus 4.6 delivers improved reasoning depth, better long-horizon task completion, and a 1-million-token context window that makes it substantially more capable for large-scale codebases and extended agentic sessions.

For teams with workloads sensitive to model behavior changes, pinning a model version in settings.json or via modelOverrides (also introduced in this release) remains the recommended approach. The new modelOverrides setting provides a clean migration path for organizations that want to route different teams or environments to different Opus versions during the transition.

/output-style Deprecated in Favor of /config

The same release deprecates the /output-style slash command. Output style is now a session-start setting — it can be configured through /config before beginning a session, but is fixed for the duration of the session once work begins. This change is intentional: locking output style at session start allows Claude Code to take advantage of prompt caching more aggressively, since a stable system prompt structure produces better cache hit rates and reduces per-turn costs.

Users who relied on /output-style mid-session should migrate to setting their preferred style in /config before starting new sessions.

Up-Arrow Interrupt Recovery

A usability improvement also shipped in 2.1.73: pressing the up-arrow key after interrupting Claude now both restores the interrupted prompt text and rewinds the conversation state in a single step. Previously, these were two separate actions. The combined behavior reduces friction when correcting or redirecting a running query.