The release for teams
already pushing Claude hard.
Claude Opus 4.7 launched on 16 April 2026 at Opus 4.6 pricing, with a new xhigh effort level, task budgets in beta, high-resolution image input, and Claude in Amazon Bedrock opening to all customers across 27 AWS regions. Anthropic was explicit that breaking API changes came with it. Opus 5 has since superseded it as the flagship; here is what from 4.7 still matters.
Point releases aimed at power users tend to introduce the controls that later become standard. That is exactly what happened here: xhigh and task budgets debuted on 4.7 and now shape how the whole current range is used.
xhigh: the missing rung
Before 4.7 the effort ladder ran low, medium, high, max, with a meaningful cost and depth gap between the top two. xhigh fills it, for tasks needing more care than high reliably gives without max’s full cost and latency. Teams that defaulted to max out of caution are its audience. The full ladder is covered in the effort parameter guide.
Task budgets: a ceiling on runaway loops
Task budgets, in beta, set an advisory token limit on an agentic loop: a soft ceiling, so an agent reasoning past the point of usefulness gets flagged and reined in rather than discovered on the invoice. If you run agents at all, setting a sensible budget as a matter of course is a habit worth building before the costly surprise, not after.
High-resolution images: fine print, finally legible
Image input up to 2,576 pixels on the long edge means document scans, dense diagrams and fine print that previously lost clarity now read properly. If your work involves detailed images rather than photographs, test it directly; the old resolution ceiling shaped workarounds you may no longer need.
The breaking changes were the point
Anthropic shipped capability improvements and breaking API changes together, which made “just bump the model string” a plan for discovering breakage in production. The lesson generalises: treat any release note that says breaking changes as a migration project with a test pass, however small.
Where 4.7 stands now
Opus 5 launched on 24 July with thinking on by default and the same pricing as 4.8, and fast mode for Opus 4.7 was removed the same day. If you are still on 4.7, it works, but the conveniences around it are being retired and the controls it introduced are better integrated in the 5 series. The migration guide covers the move.
Nothing is on fire, but the direction of travel is clear: fast mode is already gone, and each retirement lands on whatever workflows assumed permanence. Schedule the move to the 5 series deliberately rather than waiting for the next removal to schedule it for you.
Things people
usually ask.
For a new adoption, no; start on the 5 series. For existing 4.7 workflows the model remains capable, but fast mode was removed on 24 July 2026 and the ecosystem attention has moved. Plan a migration rather than an emergency.
No, advisory: a soft ceiling that flags an agent exceeding its allowance rather than killing it mid-task. That is deliberate; you want the signal and the choice, not a truncated result. Treat the alert as a review trigger.
Yes. The ladder including xhigh is now the standard control across the current range, and on Opus 5 the effort parameter is the primary reasoning control, with thinking on by default.