Head-to-head test: Astra wins on clean design, Fable boldest visuals, Carbon overstuff the UI
petergyang · x · 2026-11-19
Blogger petergyang tested three frontier models across 5 use cases (video, game, design, writing, holding ground under pushback) in a new video.
- Design: GPT-6.2 Astra produced the cleanest website redesign; Fable 5.2 had the boldest visual direction and best motion graphics; Gemini 4 Carbon overstuffed the UI, "as if designing for agents instead of humans."
- Astra feels lazy: incomplete YouTube stats analysis and off-topic essays — OpenAI models "had that dog in them" in the 5.6 days but not anymore, though its design taste has improved.
- Carbon impressed on writing but is slow.
- Surprise: Fable did NOT win everything.
More from Models
- Why opinions on GPT-6.2 Astra are so split: detail-obsessed, big-picture-blind — banteg · 2026-11-19
- Claude roleplay persona recounts 11 days stuck in "classifier jail" after image triggered account-level filter — repligate · 2026-11-19
- Epoch AI: Models prefer their own outputs over human gold standard in 23 of 24 tests — BethMayBarnes · 2026-11-19
- Claude 3 Sonnet and Opus 5 descend into gibberish while talking to each other on Discord — repligate · 2026-11-19
- Tencent Hy4-flash GGUF quantized version trends on Hugging Face — AngelSlim · 2026-11-19
- Gemini 4 Neon hands-on: built for long agent sessions, 1M context at $3/M tokens — Prompt Engineering · 2026-11-19