Anthropic Releases Claude Opus 4.7 With Reduced Cyber Capabilities

Anthropic has released Claude Opus 4.7, which the company calls its most powerful "generally available" model, though it acknowledges the model does not advance its "capability frontier." According to Anthropic's system card, Opus 4.7 received lower scores than the company's Mythos Preview model on every relevant evaluation.

Ostentatiously, Anthropic designed Opus 4.7 with intentionally limited cybersecurity capabilities. The company stated it "experimented with efforts to differentially reduce these capabilities" during training, making Opus 4.7 a testbed for new cyber safeguards before broadly releasing more powerful models like Mythos Preview.

Technical Specifications and Pricing

Anthropic claims Opus 4.7 represents an improvement over Opus 4.6 in several areas:

• Advanced software engineering tasks, particularly complex coding that previously required more guidance
• Image analysis capabilities
• Instruction following
• Creative tasks like generating slides and documents

Pricing remains unchanged from Opus 4.6: $5 per million input tokens and $25 per million output tokens.

Mythos Preview Remains Limited

Mythos Preview, which Anthropic announced earlier this month as its most powerful model overall, remains available only to select partners including Nvidia, JPMorgan Chase, Google, Apple, and Microsoft. The company cited security concerns for the limited release.

Anthropic is offering security professionals access to Opus 4.7's full capabilities through a new Cyber Verification Program, which would reduce some of the model's default safeguards for verified vulnerability research purposes.

Early Access and Testing

Early testers for Opus 4.7 included Intuit, Harvey, Replit, Cursor, Notion, Shopify, Vercel, and Databricks, according to Anthropic.

What This Means

Anthropic is taking an unusual approach by releasing a deliberately weakened version of its technology while more capable models remain under wraps. This strategy suggests growing industry concern about AI models' potential for offensive cybersecurity applications. The company is essentially using Opus 4.7 as a real-world test case for safety measures it plans to apply to future releases, signaling that access to frontier AI capabilities may increasingly depend on user verification and purpose-based restrictions rather than open availability.