Anthropic Expands AI Capabilities With Opus 4.7 Launch

Anthropic accelerates its strategy with the imminent launch of Claude Opus 4.7, a model accompanied by a tool capable of generating complete digital products from simple instructions. This advance marks a clear repositioning towards automated production platforms. In parallel, the existence of a more advanced internal model with sensitive capabilities raises questions about the uses and limits of these systems.

Anthropic attacks the creative tools market with Opus 4.7

Anthropic is about to take a big step with the launch of Claude Opus 4.7, accompanied by an automated creation tool aimed at transforming digital production. This launch could happen soon, with a clear ambition: to allow any user to generate complete products from simple instructions.

The tool would notably enable the creation of websites, presentations, or landing pages in natural language, without advanced technical skills.

The key elements of this announcement reveal an offensive strategy :

• A full-stack design tool capable of producing complete interfaces via prompts ;
• A broad target audience including both developers and non-technical users ;
• Direct competition with players like Adobe, Wix, or Figma ;
• Anthropic’s desire to embed itself into a logic of automated production of digital products.

This approach reflects a shift towards AI capable of replacing some functions traditionally handled by specialized software. By centralizing generation, structuring, and deployment, Anthropic aims to radically simplify creation processes, while redefining standards for no-code and AI-assisted design.

Mythos, an internal model with controversial capabilities

In parallel to this offensive, Anthropic is developing internally a much more advanced model named Claude Mythos, which is not intended for public use. This system was tested by the UK AI Security Institute as part of a network attack simulation called “The Last Ones”.

During this exercise, Mythos succeeded in 3 attempts out of 10, averaging 22 steps out of 32, compared to 16 steps for Opus 4.6. These results and on-chain data demonstrate a high level of autonomy in executing complex scenarios, including coordinated actions over multiple phases of an attack.

These performances underline a significant evolution of AI models towards full operational capabilities, beyond mere language processing. Mythos illustrates this transition towards systems capable of acting autonomously in sensitive technical environments.

At the same time, this progress reveals current limits of evaluation tools. Several industry players, including OpenAI, dispute the reliability of certain benchmarks, which are contaminated or not representative. Tests like ARC-AGI-3 also show considerable gaps between model performances, Gemini at 0.37%, GPT-5.4 at 0.26%, and humans, who reach 100%.

These elements reflect a profound transformation of the industry. AI is no longer limited to generating content or code. It is moving towards systems capable of designing, executing, and potentially exploiting digital environments. This convergence between creation and action opens new perspectives, while raising questions about regulation, security, and the use of these technologies as they gain autonomy.