The video evaluates various AI coding assistants’ plan modes, highlighting Cursor’s Sonic 4.5 plan mode as the most effective due to its detailed, functional, and fast project planning capabilities, outperforming Claude Code and Droid. Despite minor flaws, Cursor’s approach offers a powerful balance of thoroughness and speed, making it a valuable tool for developers.
The video discusses the evolution and effectiveness of plan modes in various AI coding assistants, with a particular focus on Cursor’s recent introduction of plan mode. The presenter reflects on their experience with plan modes, noting that while plan mode has existed for a long time, its adoption and quality have varied across platforms. They emphasize that with the latest version of Claude Code (Sonic 4.5), even simple coding tasks benefit from starting in plan mode due to the model’s limitations in edit mode, which previously filled in contextual gaps more effectively.
The presenter tests plan modes across several AI models, starting with Droid Sonic 4.5, which they found underwhelming. Despite a well-formatted basic specification, the implementation was poor, with issues like non-loading images, non-functional apps, and placeholder files. Similarly, Claude Code’s plan mode showed some improvements, such as better styling and partial functionality, but still fell short in backend API integration and overall completeness, earning a modest grade of around C minus.
Cursor’s GPT-5 High plan mode impressed the presenter with its clean full-page plan view and thoughtful follow-up questions that helped clarify project requirements. While the implementation was solid, with images loading and chat streaming partially functional, it lacked detailed file-level instructions, making it less comprehensive than desired. Despite some minor issues like unexpected dark mode formatting, this version represented a significant step forward in plan mode quality.
The standout performance came from Cursor using Sonic 4.5 with the “brain” icon enabled, which the presenter believes might be leveraging a new “Cheetah” model. This setup produced an exceptional plan with detailed API calls and file responsibilities, requiring minimal iteration. The resulting implementation was highly functional, featuring working agent selectors, chat, styling, and responsive layouts. Although not perfect—such as unnecessary modifications to SQL migrations and excessive documentation—the speed and thoroughness of this plan mode were remarkable, enabling rapid prototyping and iteration.
In conclusion, the presenter praises Cursor’s plan mode as one of the best they’ve encountered, surpassing both Claude Code and Droid in recent tests. They highlight the balance between speed and thoroughness, noting that while Cursor’s Sonic 4.5 plan mode isn’t flawless, it offers a powerful tool for developers seeking efficient and detailed project planning. The video ends with an invitation for viewers to share their experiences and opinions, acknowledging that while some may disagree, the presenter’s testing strongly favors Cursor’s approach, especially when paired with Sonic 4.5 over GPT-5 High.