Sonnet 5 review: I ran 64 generations to find out if it's worth it

Short summary

Lenny benchmarked 5 frontier AI models through 64 blind prototype generations, testing how each handles two practical professional tasks: analyzing PRDs and managing agent voice interactions, with Claude Code as the evaluation platform. Sonnet 5 delivered unexpectedly strong performance against comparable models. The full benchmark methodology, detailed performance metrics for each model, and comparative analysis are available in the video.

• •Blind benchmark tested 5 frontier models across 64 prototype generations
• •Evaluated PRD analysis and agent voice task performance using Claude Code
• •Sonnet 5 showed surprisingly strong results (full details in video)

Generated with AI, which can make mistakes.