Anthropic introduced its latest model, Claude Opus 4.7, on Thursday, noting that it is not as powerful as its limited-release cybersecurity model, Claude Mythos.
Opus 4.7 is more advanced than Opus 4.6 in software engineering, with early users being able to hand off some coding work to the model without oversight, Anthropic said. Opus 4.7 can handle longer-running tasks and pay closer attention to instructions. It is also better at interpreting images and can be more creative when creating slides and documents.
While Opus 4.7 is not as powerful as Claude Mythos — which triggered a tech-industry scare because of its ability to easily hack software — Anthropic said security professionals can still use it for cybersecurity tasks such as vulnerability research, penetration testing and red teaming.
“This is a lot more about Opus finding its place within the enterprise for actual business tasks, difficult business tasks, like migrating a database from one platform to another,” said Bradley Shimmin, an analyst at Futurum Group.
With Opus 4.7, Anthropic aims to address challenges enterprises face when working with frontier models. For example, the vendor included features such as loop resistance, which prevents the model from getting into a loop that requires the human user to start over, and even pushes back against hallucinations within the model. These features make enterprise experiences with models much smoother, the generative AI vendor claimed.
Addressing Enterprise Problems
Anthropic is also being more intentional about making its model more than just another frontier model, forming it into a platform and harness that supports enterprises, Shimmin said.
Capabilities such as how a model manages basic commands or skills, or how it accesses the file system or memory, are not necessarily part of the model but determine how it performs. New features in Opus 4.7, such as auto mode, show the vendor is focusing on providing enterprises with a model that performs more like a platform, with a rich API and a wide range of methodologies that set it apart from frontier models from competitors such as OpenAI and Google, Shimmin added.
A key improvement Anthropic made with Opus 4.7 is in agentic reasoning capabilities, said Arun Chandrasekaran, an analyst at Gartner.
“They have much better benchmarks in terms of coding tasks, which include everything from refactoring, [updating] legacy code and documentation of code,” Chandrasekaran said. “All of these are domains within software development where the 4.7 model shows much better results.”
The Need for Security
However, even with the improved capabilities of Opus 4.7, the model is still not as powerful as Claude Mythos in preview.
For Shimmin, it seems that Anthropic is walking the line, recognizing that as AI models become more advanced, users need agentic AI to be secured and also to provide security.
“It’s like this multi-faceted sort of situation where it opens up a lot of attack vectors, but it can be used to solve a lot of attack vectors,” he said.
