Anthropic releases Claude Opus 4.7

Background

Anthropic released Claude Opus 4.7 as generally available on 16 April 2026, at $5 per million input tokens and $25 per million output tokens. ^[1]

What happened

Anthropic said Opus 4.7 was less broadly capable than Claude Mythos Preview, and that its cyber capabilities had been deliberately reduced. ^[1][2]

Opus 4.7 was the first model to test automatic cyber safeguards, ahead of a broader release of Mythos-class models. ^[1]

Key details

A new effort level, xhigh, became the default in Claude Code. ^[1]

The model accepts images up to 2,576 pixels on the long edge, which Anthropic said was more than three times the limit of earlier Claude models. ^[1]

Claude Code added /ultrareview, a review command, and extended auto mode to Max users. ^[1]

Automatic cyber safeguards are rules that detect and block risky cyber requests without a person reviewing each one. Reduced cyber capability means that a model is intentionally weaker in areas where its abilities could be misused. Both are ways a company can limit how a model is used before it releases a more capable one more widely.

Impact

Anthropic described Opus 4.7 as a step in testing automatic cyber safeguards ahead of a broader release of Mythos-class models. ^[1]