Vibe Check: Claude 4 Opus

Dan Shipper reviews Anthropic's Claude 4 Opus, concluding that it is a strong model for editing, coding and long research tasks, though he still uses OpenAI's o3 for everyday work. He says Opus edits honestly, judging writing against a set of principles he and Danny Aziz provided, and it tracked those principles across long prompts where other reasoning models lost them. In Claude Code it built a feature for Cora that was close to shippable, and its research mode spawned parallel agents that gathered 645 sources for one query. Shipper adds that model performance is now good enough that user interfaces and features like memory matter more, and that ChatGPT's memory keeps him on it.