This is an actual f*cking goldmine
Anthropic just shared one of the best tricks for Claude Code. The idea: run Sonnet, Haiku 5.5 and Opus 5.5 at once, each one doing only the job it's best at. Better results and way fewer tokens, because every task lands on the right model.
Fable 5.2 launched today.
My feed says "best model ever", so I tested it against Opus 5.5. 17 coding tasks. 68 scored runs. 5 tasks were real commits from my production app. Same quality, about half the API cost. Numbers and videos below 👇
This is not Fable 5.2! Everything you see (aside from the Tesla cyber truck) was one shoted with Opus 5.5 before it got nerfed.
My prompt was very simple: Remake GTA2 game with modern graphics, everything made with code. It worked for 5H and then I was blown away by the result.
This is f*cking insane. Anthropic engineer Andrej Karpathy
"The biggest mistake in AI right now is forcing agents to work before you master the harness around the model We made that mistake at OpenAI in 2016. It cost us 5 years" In just 6 minutes, he explains why the agent is
You have reached the end of the archive
All of Claude Opus 5