Opus 5.5 is insanely powerful In 24 hours,
solo-developed a fully original playable game
Tech stack: Opus 5.5 + Flux 3 Max + GPT Image + Tripo + Blender + Three.js Making games is another option
WAIT, THIS WAS BUILT WITH MINIMAX M3.1 FLASH??? ๐ญ
someone used the model + Three.js to create an entire explorable medieval village
cozy lighting, detailed little houses, and an atmosphere that looks straight out of an indie game
took a few iterations, but the result is beautiful
AI-generated 3D worlds are getting seriously interesting.
FIRST "LE GPT FAT"
NOW "CHATBLT" ๐ญ
Tibo is either dropping the most cryptic hints about Betelgeuse...
or he's literally planning his lunch while the entire AI timeline tries to decode it ๐
ARE WE GETTING A NEW GPT OR A SANDWICH???
SOMEONE ASKED FABLE 5.5 TO SHOW WHAT KINDNESS LOOKS LIKE.
and it came up with this ๐ฅน
a little character gives its light to a child, only for its own light to return even brighter
the ending is genuinely beautiful
and apparently every frame AND the sound were generated through code
AI is getting ridiculous.
this is actually insane
one sentence in and both AIs took me straight into the same nightmare
a desert carpet ,a serpent swallowing the frame then a fall into something dark and alive
no cuts , no edits just one unbroken shot from start to finish
top: Gemini 4 Argon. bottom: Opus 5.5
same prompt completely different fever dreams a year ago these models were melting faces and giving people 6 fingersbnow they're doing this ,Gemini or Opus
which nightmare are you watching twice?
Unbelievable ๐ฑ, Someone used Claude Opus 5.5 + Three.js to build a ring-shaped structure floating above a sea of clouds in the snowy mountains
I really thought it was a video generated by a video model, but turns out it's a 3D scene built with code!!! ๐คฏ
CLAUDE OPUS 5.5 HAS A STOPPING PROBLEM ๐ญ
it completed a robot-arm task in 23 minutes
even said: "The crocodile looks fully inside now"
then spent ANOTHER 22 MINUTES pushing the crocodile back and forth...
only to end up almost exactly where it was when it first finished ๐
apparently knowing when to stop is a skill too
unpopular opinion: a coding agent passing all your tests doesn't mean it wrote correct code
it means it satisfied the checks you gave it
Claude, Codex, Gemini, doesn't matter
the real question is:
how do you catch the bugs your tests never thought to look for?
serious question for devs using Claude Code and Codex daily
are you genuinely shipping better software now, or just shipping MORE software?
there's a difference and i'm curious how people measure it