I wanted to sticky-label my room to learn English.
But I didn't. And I wanted a real "English Street" in my city. But there isn't one. So I used Opus 5 claudeai & OpenAI GPT-5.6 Astra sama to build one in 1w. keke: a low-poly 3D town to tap, speak, and learn in real scenes.
Used Blender MCP with Claude code and made this animation with back and forth conversation and tweaking with Claude!
This whole session used about 33.3 million tokens across 132 requests to Claude, all on Opus 5. At Anthropic's standard API prices that would come to roughly $42.
One of the most powerful flows in agentic coding that people are not utilising is mixing harnesses and models.
Here's a demo of GPT-6 Astra in Codex communicating through tmux-ide with Claude Code. Different harnesses & models excel at different goals. E.g Fable 5.1 is still
Opus 5.2 is coming AT THE END of this MONTH
Polymarket gives an 81% chance of a September release the model is being tested as Opus 5 in Claude Code prompt to check your CC in the comments Opus 5.2 isn't lazy; it handles details very well it has a great conversational style
You have reached the end of the archive
All of Claude Opus 5