My first test with Gemini 4 pro (under gemini 3.7 flash in arena)
I think it beats Opus 5 Prompt: Create a 3D model of an Ak-47 in Three.js
I gave one brief to two coding agents and walked away.
Same data, same page. One bill was 40x the other. The brief: take 1,038 signals from my crypto judge with the price 24h later, build a page that shows how often each score bucket rugged. Claude Code + Opus 5: 6 min, $1.19.
I've been thinking about this all week, and I'm now 100% convinced OpenAI fumbled the developer ball.
When GPT-5.6 Sol launched, the resets were generous. That let people build what I think of as model sense: what this model is good for and what it isn't. It also
Love this design from gasphedde 🖤
Tried to rebuild it with a single prompt using Claude Opus 5. 3D turntable, 30 real records.
You have reached the end of the archive
All of Claude Opus 5