Anthropic Preps Opus 4.7 and Full-Stack AI Studio—While Sitting on Something Much Scarier

Anthropic is gearing up to merchandise Claude Opus 4.7 alongside a caller AI-powered plan instrumentality that lets users physique websites, presentations, and landing pages with plain English prompts—news that caused a dip successful Adobe, Wix, and Figma shares connected Monday, according to The Information.

The products could driblet arsenic soon arsenic this week, a idiosyncratic with cognition of the plans told The Information. The plan instrumentality targets developers and non-technical users alike, putting it connected a collision people with startups similar Gamma and Google's Stitch.

Anthropic did not respond to Decrypt’s petition for comment.

Opus 4.7 isn't adjacent Anthropic's astir almighty model. That rubric belongs to Claude Mythos—a cybersecurity-focused beast the institution is softly handing to prime information firms portion keeping it distant from the public.

The UK's AI Security Institute recently evaluated Mythos Preview and recovered it tin autonomously execute blase cyber attacks astatine rates nary different exemplary has matched. It became the archetypal AI to implicit "The Last Ones," a 32-step firm web onslaught simulation that typically takes quality reddish teams 20 hours. Mythos nailed it successful 3 retired of 10 attempts, averaging 22 of 32 steps—compared to Opus 4.6's 16.

This matters beyond endeavor security. Measuring what AI tin really bash has go an industry-wide headache. OpenAI precocious called the starring coding benchmark "contaminated," yet models proceed to beryllium compared utilizing those aforesaid tests. A abstracted ARC-AGI-3 evaluation saw Gemini people 0.37% and GPT-5.4 deed 0.26%—while humans got 100%. The effect is simply a scenery wherever benchmarks are some contested and inactive utilized arsenic evidence, making it hard to contextualize claims astir Opus 4.7’s gains until Anthropic releases a elaborate exemplary card.

The narration betwixt Opus and Mythos is person than astir realize. Anthropic builds its frontier models by fine-tuning atop the Opus line—the aforesaid backbone powering nationalist Claude products gets stress-tested and hardened into Mythos. Opus 4.7 is the instauration that yet gets the cybersecurity kung fu beaten into it.

Also, Anthropic’s efforts person been steering much towards the development/enterprise usage case. The leak of Claude code, the merchandise of the skills strategy and MCP protocol, the absorption connected agentic AI and the attraction connected coding benchmarks marque this adjacent much apparent. While Anthropic hasn’t formally announced it, the leaks reenforce the broader displacement from LLM supplier to thing that resembles a full-stack “AI studio” model, wherever Claude doesn’t conscionable make substance but builds and deploys implicit products.