A leaked GPT-6 Astra video is going around.
The prompt was one line: "build a gorgeous sci-fi spaceship with explorable internal structures" one shot. no iterations. it shipped deck isolation, hull transparency, a room index and a working 3D viewer. ppl thought Opus 5 could
More tx = more butterflies, I kinda wish I'd explicitly told them all to animate the Nervape like…
More tx = more butterflies, I kinda wish I'd explicitly told them all to animate the Nervape like Gemini did, having the ape chase the butterflies was a better idea than I came up with😂. Opus 5, Kimi K3, GLM 5.3 flash, Gemini 3.7 all serviceable here. Seems all AI likes threeJS.
Seven frontier models were tested.
The best score, from Claude Opus 5, was 0.325. none of them passed. This model didn't pass either. That gap, between acing a closed benchmark and surviving messy real life, is the actual problem they're trying to solve. this is their first
5.6 sol max vs claudeai opus 5 max in the game of chess.
Who will win?
You have reached the end of the archive
All of Claude Opus 5