Anthropic has released its latest model, Claude Opus 4.7, featuring a significant improvement in programming and visual capabilities, along with the integration of a state-of-the-art cybersecurity defense mechanism.
BlockBeats News, April 16th, Anthropic released its latest model, Claude Opus 4.7, achieving a significant advancement in the field of advanced software engineering compared to Opus 4.6, especially excelling in high-difficulty tasks. Users reported that high-difficulty programming tasks that previously required close supervision can now be confidently handed over to Opus 4.7 for independent completion. The model can maintain rigor and consistency in complex long-process tasks, self-validate output results before feedback.
In terms of cybersecurity, Opus 4.7 is the first model under the Project Glasswing initiative to incorporate a new type of cybersecurity protection mechanism, capable of automatically detecting and intercepting prohibited or high-risk cybersecurity-related requests. Its cybersecurity capability is inferior to Mythos Preview. Anthropic stated that it will leverage this deployment to accumulate experience in preparation for the widespread release of Mythos-level models. In terms of visual capabilities, Opus 4.7 supports image inputs with a maximum long edge of 2,576 pixels (approximately 3.75 million pixels), three times higher than the previous Claude model, further expanding the application scenarios for computer-assisted agents, complex chart data extraction, and other multimodal applications.
Opus 4.7 is now available on Claude's full product line and API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. The pricing is the same as Opus 4.6, with an API input price of $5 per million tokens and an output price of $25 per million tokens. Developers can make calls through claude-opus-4-7. This update simultaneously introduces several new features: the addition of an xhigh effort level (located between high and max); the task budget function is launched on the API side in the form of public testing; Claude Code adds the /ultrareview deep code review command and extends automatic mode to Max users.
