Opus beats Astra on one-shot 3D quality; Astra is 4x faster and 10x more token-efficient
chaseleantj · x · 2026-09-13
In a follow-up to his 3D scene generation tests, the author confirms recent reports that Astra's quality has dropped: even at high effort with a skill and image reference, Astra took 14 minutes and 14k output tokens and still fell clearly short of Opus.
He notes Astra was 4x faster and >10x more token-efficient, meaning its effective ceiling may be higher when pushed — he plans more tests. This experiment only compared one-shot quality, where Opus clearly wins.
More from Models
- Dev burns Claude's 5-hour limit in one hour, switches to Codex for implementation — rudrank · 2026-09-13
- Hugging Bay: unofficial index of ~149k open AI models, weights, datasets and tools — johnseach · 2026-09-13
- Distillation means a frontier pause still pushes Astra-grade models to $0.3/1M tokens — teortaxesTex · 2026-09-13
- GPT-6 Astra one-shots a full portfolio website in a single prompt — Tegadesigns · 2026-09-13
- Hugging Bay indexes 149k public open-source AI models with licenses and SHA-256 hashes — johnseach · 2026-09-13
- Doubao stands out in Arena model-similarity analysis, sparking distillation debate — teortaxesTex · 2026-09-13