In its latest weekly briefing covering April 12–18, 2026, Anthropic announced the release of Claude Opus 4.7, achieving an 87.6% score on the SWE-bench Verified benchmark — a new record for the platform. The model also features 3.75 MP vision limits, enabling more detailed image analysis.
Alongside the model update, Anthropic Labs introduced Claude Design, a new tool for AI-driven visual prototyping and UI mockups, expanding Claude's capabilities beyond text and code into design.
The briefing also highlighted progress in AI alignment research: Automated Alignment Researchers achieved a 0.97 PGR score, surpassing human baselines. In a strategic move, Novartis CEO Vas Narasimhan joined Anthropic's board, securing a Long-Term Benefit Trust majority. Meanwhile, the UK AISI evaluated Claude Mythos Preview on 32-step cyberattack simulations, testing the model's safety in adversarial scenarios.