Claude 4 models are out…. sort of. Here’s what’s going on when you try to use it with some external tools:

What’s Awesome

• Understands your intent and picks the perfect tool
• Pipes data between services without a sweat
• Wraps it all into neat, readable answers

What’s a Buzzkill

• Mid-stream dropouts: Responses randomly cut off.
• 429 errors: Even light usage occasionally hits “Too Many Requests.”

Claude 4 models have been overran by demand. Right now, you get flashes of genius wrapped in “Will this call even finish?” tension.

Cheers to a tool with promise—just needs more reliable delivery!

ps: just in case you want to see what’s happening behind the scene, with some logging info about tool use.