Anthropic: Claude Opus 4.7 in the Shadow of Mythos Preview

Anthropic recently unveiled Claude Opus 4.7, touted as the company's most powerful model available to the public. This model represents a significant advancement over its predecessor, Opus 4.6, in terms of software engineering. It is particularly effective at handling complex coding tasks that previously required increased assistance. Additionally, Opus 4.7 enhances image analysis and instruction tracking, while offering improved creativity in creating slides and documents.

However, Opus 4.7 arrives in the shadow of Mythos Preview, a cybersecurity-focused model announced by Anthropic earlier this month. Mythos Preview is considered the company's most powerful model to date, surpassing Opus 4.7 in every relevant evaluation. In fact, Opus 4.7 does not even push the "capabilities frontier" of the company. For security reasons, Mythos Preview is currently accessible only to selected partners, such as Nvidia, JPMorgan Chase, Google, Apple, and Microsoft.

Anthropic clarified that the release of Mythos Preview would remain limited, allowing for the testing of new cybersecurity protections on less powerful models like Opus 4.7. During the training of Opus 4.7, the company experimented with efforts to differently reduce cybersecurity capabilities compared to Mythos Preview. This approach aims to strengthen cybersecurity protections while paving the way for a broader rollout of Mythos-class models.

Early users of Opus 4.7 include companies such as Intuit, Harvey, Replit, Cursor, Notion, Shopify, Vercel, and Databricks. Anthropic confirmed that the pricing for Opus 4.7 remains the same as that of Opus 4.6, at $5 per million input tokens and $25 per million output tokens. Security professionals looking to leverage Opus 4.7 for vulnerability research can join Anthropic's Cyber Verification Program, which allows for the lifting of certain protections put in place for this model.