Claude Sonnet 5.5 vs GPT-6 Astra: 3D city test on the $20 tier
On the same task, Sonnet 5.5 produced more detail and a city four times larger but used 68.7 million tokens, while Astra used only 1.62 million, and the author prefers the pre-nerf Astra.
🚨 codex and claude users!! I tested the $20 Claude Pro Sonnet 5.5 against GPT-6 Astra!! First off, the usage: - 68.7 Million tokens (INSANE!!!) - 2 hrs and 44 minutes work - 19% weekly usage (142% 5h limit) - $29.48 of API usage Now... look at the QUALITY!! MUCH MORE attention to detail and 4x large city by Sonnet 5.5 (No. of frames were equal, GPT-6 Astra exported in 60 FPS) Absolutly beautiful work done by BOTH BUT I'd prefer GPT-6 Astra (pre-nerf) since it only consumed 1.62 Million tokens for about 15% weekly usage. My recommendation: If you want more "dictated" work, choose Claude $20... If you don't wanna lead and just let it do the stuff... Go with Codex $20 (for now) NOTE: I used BOTH on max effort to get the absolute best results.