GPT-6-Astra | First impressions

https://arena.ai/agent GPT-6 Astra just dropped from OpenAI, and Peter puts it through an extensive, zero-cherry-picking gauntlet of one-shot 3D and interactive world generations. We compare Astra directly against Fable, while also breaking down how reasoning level (low through ultra) changes output richness and whether the extra cost is worth it. The verdict: after several generations where GPT models trailed Fable, Astra is now genuinely competitive — both models are excellent, each with its own sharp edges, and it's a good moment to stop assuming last generation's pecking order still holds. Peter’s prompts: https://github.com/petergpt/3d-prompt-collection 00:00 – Intro: GPT-6 Astra has landed 00:19 – Voxel London through the ages 01:39 – D-Day: Omaha Beach in Blender + Three.js 03:25 – Open-world exploration game (10-hour task) 05:36 – One-shot generation gauntlet begins (max reasoning) 07:15 – Neuschwanstein-style castle 09:06 – X-ray transition scene & underwater fish world 10:48 – Zooming into a leaf 11:20 – Pyramid construction 11:58 – Impressionist painting world 12:56 – Van Gogh's house walkthrough 14:48 – Great Fire of London / the Blitz 20:04 – Humpback whale 20:37 – Swiss Alps trail (research-based prompt) 23:05 – Reasoning level breakdown: low vs. medium vs. high vs. extra high vs. max vs. ultra 28:51 – Final head-to-head: Westminster, White House, dinosaurs 31:00 – Verdict: is Astra finally on par with Fable? #arenaai #openai #gpt #llm