TLDR Founder Dan Says Opus 5.5 Beats Opus 5 on Personality Not Benchmarks
Summary
TLDR founder dan argues that model personality and writing quality outweigh benchmark scores in how users judge AI, pointing to Opus 5 scoring well but drawing dislike while Opus 5.5 is widely preferred. Commenters split, with some saying verified output matters more when agents run coding tasks for hours.
Key Points
- dan says his personal evals now test mostly writing, checking whether a model states the interesting point clearly and succinctly.
- photonstorm, CTO at Phaser Studio, says Gemini writes best but Claude remains his daily coding driver despite its tone.
- node, AI R&D in Manufacturing, says personality matters less when dev agents run for hours and wants more evidence.