FULL STORY

Four Models Benchmarked on 3D Generation: Opus Quality vs Astra Speed

A developer benchmarked four models on the same 3D generation task, finding Opus leads in design quality while Astra is faster and cheaper but shows quality decline, concluding that skill-heavy one-shot generation beats iterative fixing.

2026-09-13 ~ 2026-09-13 · 4 episodes · 8 posts

Episode 1 · Four-Model 3D Test: Opus Leads in Design but Costs 3x Astra's Time and Tokens (2026-09-13, 2 posts)

A four-model comparison on the same 3D dandelion prompt ranks Opus first, with sub-high effort deemed not worthwhile; Opus's design beats Astra at roughly 3x the time and token cost.

Episode 2 · AI Can't One-Shot Polished 3D: Manual Iteration and Reference Images Are Key (2026-09-13, 2 posts)

The author finds no model can one-shot a polished 3D result, requiring manual curation and a 17-minute merge, while adding reference images improved detail but raised output from 153k to 252k tokens.

Episode 3 · 3D Generation Tests: Upfront Techniques Beat Iterative Fixes as Astra Quality Slips (2026-09-13, 2 posts)

The author concludes that skill- and reference-based one-shot generation beats iterative fixing on both quality and token cost, and confirms community reports that Astra's quality has declined versus Opus.

Episode 4 · Hands-on: Opus Beats Astra in Quality, but Astra Is 4x Faster and 10x More Token-Efficient (2026-09-13, 2 posts)

Hands-on tests show Astra's quality has declined and still lags Opus even at high effort with skills and reference images, though it runs 4x faster and is over 10x more token-efficient.