Claude Design's Replication Ability Degrades
ivan_bezdomny · x · 2026-07-16
The author reports that it has become extremely difficult for Claude Design to reproduce designs it created just a month ago. Even when placing the original design files into the context for the model to read, stable replication remains a struggle.
The poster believes this change feels like the model has "gotten dumber," or perhaps tasks now need to be bumped to the max tier, meaning the past high setting has effectively become the new low.
Related event: Users Complain About Claude's Declining Speed and Quality(4 posts)→
More from Models
- Benchmark scores drop from 89% to 19% on new evals — how benchmaxxing breaks leaderboard trust — airesearch12 · 2026-09-11
- ChatGPT tells user their question is too hard and to 'accept dumber answers' — phido3000 · 2026-09-11
- Developer Building a Unified Leaderboard of All Model Benchmark Scores — airesearch12 · 2026-09-11
- Rumor claims Kimi faked performance by serving Claude; DeepSeek new model surprises in evals — realsohamparekh · 2026-09-11
- GPT-5.6 writes well but is instantly forgettable, user complains — BasedRaddka · 2026-09-11
- Opus Refuses Protein Research Codebase Over 'Safety' Concerns, Dev Considers Rolling His Own — josephdviviano · 2026-09-11