Claude Opus 4.7 Arrives in Claude Code: xhigh Effort, Auto Mode for Max, and /ultrareview

Claude CodeView original changelog

Claude Code version 2.1.111 ships full support for Claude Opus 4.7, Anthropic's most capable generally available model to date. The release introduces a new xhigh effort tier between high and max, an interactive /effort slider for real-time tuning, and Auto mode now unlocked for Max subscribers using Opus 4.7 without requiring any flag. A new /ultrareview command enables comprehensive cloud-based code review across parallel agents, and plan files are now named after the user's prompt for easier multi-session navigation.

Key Takeaways

  • Claude Opus 4.7 is now the default model in Claude Code, posting 87.6% on SWE-bench Verified and a 13% improvement on coding benchmarks over Opus 4.6.
  • The new xhigh effort tier β€” sitting between high and max β€” is the recommended default for agentic coding, delivering most of max-level reasoning at lower latency and cost.
  • Auto mode no longer requires a flag and is now available to Max, Teams, and Enterprise subscribers, enabling fully autonomous permission handling for safe commands.
  • /ultrareview launches parallel cloud agents for exhaustive background code review, with Pro and Max subscribers receiving three free trials.
  • /less-permission-prompts scans transcripts to auto-generate a safe-command allowlist, reducing repetitive approval interruptions in daily coding workflows.
  • Plan files are now named after the user's prompt, improving session navigation when multiple plans are active simultaneously.

Claude Opus 4.7 Comes to Claude Code

Claude Code version 2.1.111 marks a significant milestone: full integration of Claude Opus 4.7, Anthropic's strongest generally available model to date. Released on April 16, 2026, this update reshapes how developers interact with the tool β€” from effort control to autonomous approval flows β€” building on a model that posts 87.6% on SWE-bench Verified and 64.3% on SWE-bench Pro.

A New Effort Tier: xhigh

Anthropic introduced a new xhigh effort level that sits between high and max, giving developers finer-grained control over the reasoning-to-latency tradeoff on complex tasks. The /effort command now opens an interactive slider when called without arguments, allowing real-time tuning without leaving the session. The default effort for Opus 4.7 in Claude Code is xhigh, which Anthropic recommends for most coding and agentic work β€” it delivers the bulk of max-level reasoning at substantially lower latency and cost.

For teams managing token budgets across concurrent sessions, high remains the cost-effective choice, while max is reserved for genuinely hard problems where the additional reasoning cycles justify the overhead.

Auto Mode Opens to Max Subscribers

Auto mode β€” which routes permission prompts through a classifier that determines whether a proposed action is safe to approve silently β€” is now available for Max subscribers when using Opus 4.7. It no longer requires the --enable-auto-mode flag, reducing friction for users who want fully autonomous agentic workflows. Teams and Enterprise subscribers also gain access. Safe commands proceed without interruption; riskier operations still surface a prompt as expected.

/ultrareview: Cloud-Based Comprehensive Code Review

A new /ultrareview command launches a cloud-based multi-agent code review against a pull request or file set. Pro and Max subscribers receive three free trials. The command is designed for thorough review passes β€” more exhaustive than an in-session review β€” and runs in the background so developers can continue working while the review completes.

/less-permission-prompts: Reducing CLI Friction

The new /less-permission-prompts skill scans recent session transcripts and proposes an allowlist of commonly approved, read-only commands. Accepted rules are written to the project's settings so those commands auto-approve in future sessions β€” a meaningful quality-of-life improvement for developers who run the same diagnostic operations repeatedly.

Quality-of-Life Improvements

Several smaller but meaningful changes ship alongside the Opus 4.7 integration:

  • Plan file naming: Plan files are now named after the user's prompt (e.g., fix-auth-race-snug-otter.md) rather than generic filenames, making multi-task sessions easier to navigate.
  • Auto theme: A new "Auto (match terminal)" theme option dynamically adapts Claude Code's color scheme to the terminal's dark or light mode setting.
  • Ctrl+U / Ctrl+Y: Ctrl+U now clears the entire input buffer instantly; Ctrl+Y restores it β€” handy for discarding a long prompt without losing it permanently.
  • Typo recovery: Claude Code now suggests the closest matching subcommand when a claude <word> invocation has a near-miss typo, reducing friction for new users.
  • Glob patterns in read-only commands: Read-only bash commands that use glob patterns no longer trigger permission prompts, eliminating a common source of interruption.

Bug Fixes

This release resolves a number of notable issues:

  • Fixed terminal display tearing in iTerm2 + tmux setups
  • Fixed @ file suggestions re-scanning the entire project on every turn
  • Fixed /clear dropping the session name set by /rename
  • Fixed tab-completing /resume immediately resuming an arbitrary titled session
  • Fixed LSP diagnostics from before an edit appearing after it
  • Fixed 429 rate-limit errors on Bedrock/Vertex/Foundry referencing the wrong status URL
  • Fixed feedback surveys appearing back-to-back
  • Fixed bare URLs in bash/PowerShell/MCP tool output being unclickable when wrapped