Anthropic drops Claude Opus 4.7

This is a meaningful refinement release, not just a bigger model number: Anthropic is tightening the knobs that matter for serious agent workflows, especially latency control, token spend, and review quality.

• –`xhigh` gives developers a new middle ground between `high` and `max`, which is exactly the kind of control teams want when hard tasks start to get expensive.
• –Task budgets are the more interesting API addition for production use because they turn token usage into something you can steer over long runs instead of merely observe after the fact.
• –Better vision at higher resolution should improve Claude’s usefulness in screenshot-heavy agent tasks, docs, and slide work, where tiny visual details often make or break the output.
• –`/ultrareview` and expanded auto mode reinforce Anthropic’s push to make Claude Code feel less like a chat tool and more like an autonomous workbench.
• –Same pricing as Opus 4.6 makes this an easy upgrade decision for teams already paying for frontier performance.