Anthropic revealed 6 core rules for prompting Claude 5 models 👇

Most of what worked 5 months ago now makes the output WORSE. 1⃣ Define what done looks like. Claude 5 models do NOT do too little, they tend to do too much. Not defining an exit criteria means it keeps going and

Pietro MontaldoSep 14

Scroll this and watch a cloud of dust turn into an ear.

There isn't a single photograph on the page. Every image you see is the same particles, rearranging the name, a voice, the inside of an ear, then back to the logo. Built with Claude Opus 5.

Ohad KrispinSep 142

He said Dario was right.

Then he numbered the next four Groks in one afternoon. 4.7 ≈ Opus 5.0. Not 5.1. 4.8 is the leap. Still training. 4.9 is Astra class. 5 is "maybe better than anything." The only Grok in Word tonight is 4.6. The roadmap is a launch. The product is a

kenokkiSep 142

DeepSeek-V4.1-Flash (Max) is a breakthrough in performance to cost efficiency.

With +4.87% net improvement at $0.07 cost per median task, it’s reshaped the Pareto frontier for Agent Arena! Among the top 3 open models, DeepSeek-V4.1-Flash (Max) has the lowest median task cost.

Arena.aiSep 14594