[DigitalToday reporter Chi-gyu Hwang (황치규)] Anthropic has launched Claude Opus 4.7, the latest model in its Opus series.

The company said Opus 4.7 is a model with strengthened software engineering capabilities, showing notable performance improvements, particularly on highly difficult coding tasks.

It features improved vision functions through support for high-resolution images, greater completeness and creativity in professional work such as interfaces, slides and documents, and stronger instruction-following capabilities.

Opus 4.7 is not a cybersecurity-focused model. In cybersecurity capabilities, it falls short of the recently announced Claude Mythos Preview, and experiments were also conducted in parallel to intentionally lower security capabilities during training.

With the launch, a new safeguard has been introduced that automatically detects and blocks prohibited or high-risk cybersecurity-related requests. The company explained that this is also the first case of applying in a real-world environment the approach presented in last week's Project Glasswing.

Mythos Preview is Anthropic's top-performing model and is being provided on a limited basis only to cyber defense experts and critical infrastructure partners. Opus 4.7 is Anthropic's top-performing model among those available for general access. Its pricing is the same as Opus 4.6: $5 per 1 million input tokens and $25 per 1 million output tokens.