Anthropic has introduced its most advanced “generally available” model yet: Claude Opus 4.7. The company claims it improves upon Opus 4.6, particularly in advanced software engineering tasks and complex coding, requiring less guidance. It is also designed to better analyze images, follow instructions, and demonstrate increased creativity in creating slides and documents. However, Opus 4.7 follows the release of Mythos Preview, a cybersecurity-focused model that Anthropic earlier this month referred to as its most powerful model. In comparison, Opus 4.7 is more restricted, with Anthropic stating in its system card that it does not push the company’s “capability frontier,” as Mythos Preview outperformed it in every relevant evaluation. Mythos Preview is currently available only to select partners such as Nvidia, JPMorgan Chase, Google, Apple, and Microsoft, for security reasons. Anthropic explained in a blog post that they would limit Claude Mythos Preview’s release and first test new cyber safeguards on less advanced models. Opus 4.7 is the first such model, with cyber capabilities not as advanced as Mythos Preview, as Anthropic experimented with reducing these capabilities during its training. The release of this new model includes additional cybersecurity safeguards compared to Opus 4.6, and observations from these safeguards will aid Anthropic in potentially releasing Mythos-class models more broadly. Additionally, security professionals interested in using the model for cybersecurity, such as vulnerability research, could join Anthropic’s Cyber Verification Program, which may relax some of the Opus 4.7 safeguard measures. Initial testers of Opus 4.7 included Anthropic clients such as Intuit, Harvey, Replit, Cursor, Notion, Shopify, Vercel, and Databricks, with pricing remaining the same as Opus 4.6 at $5 per million input tokens and $25 per million output tokens.
New Anthropic Opus Model Launched Amid Mythos Preview Buzz
2 Min Read