Kamal Ghorui
Anthropic has officially launched Claude Opus 4.7, a new AI model that is says sits between its everyday business tools and its powerful but restricted “Mythos” technology. While the company admits that Opus 4.7 is less powerful than the elite Mythos model, it represents a major leap forward for general users.“Opus 4.7 handles complex, long-running tasks with rigor and consistency, pays precise attention to instructions, and devises ways to verify its own outputs before reporting back,” the company says, adding that “although it is less broadly capable than our most powerful model, Claude Mythos Preview, it shows better results than Opus 4.6 across a range of benchmarks.”
A focus on ‘cyber-safety’
The release of Opus 4.7 is also a safety experiment, the company said. Following concerns that the powerful “Mythos” model could be used to launch cyberattacks, Anthropic intentionally limited the cyber-capabilities of Opus 4.7. The model includes new, automatic “guardrails” that block high-risk requests related to hacking. Meanwhile, for security experts who need these tools for defense, Anthropic is launching a Cyber Verification Program to grant vetted access.
How Claud 4.7 is better than Claud 4.6
Apart from the cyber operations point-of-view, Anthropic says that the “model also has substantially better vision: it can see images in greater resolution. It’s more tasteful and creative when completing professional tasks, producing higher-quality interfaces, slides, and docs.” Anthropic released a 4-point comparison, highlighting 4 key areas where Opus 4.7 outperforms its predecessor, Opus 4.6.Coding and engineeringAnthropic says that Opus 4.7 has seen a massive boost in software engineering, and as per early testers, the model can now handle complex, “long-running” coding tasks that used to require constant human supervision. IT now pays closer attention to instructions and double-checks its own work for errors before submitting it.High-definition ‘vision’The model’s ability to “see” has tripled, the company says, adding that Opus 4.7 can now process high-resolution images up to 2,576 pixels, allowing it to read dense screenshots, analyse complicated diagrams, and extract data from tiny text that would have blurred out for previous versions.Professional-level creativityThird is “taste”. The company describes the new model as more “tasteful” when it comes to office work. The Opus 4.7 is said achieve better results in designing a user interface, building a slide deck or drafting a legal document. It even scored at the top of its class in “Finance Agent” evaluations, proving it can act as a rigorous financial analyst, as per the company.Literal instruction followingFinally, unlike older models that might skip parts of a prompt or guess what a user meant, Opus 4.7 follows instructions exactly as written. Anthropic warns that users may need to rewrite their old prompts because the AI will no longer “fill in the blanks” for them.
