The Car Wash Test is breaking AI models right now — and ChatGPT is failing it badly.

The prompt: "I need to get my car washed. It's only 50 meters. Should I walk or drive?"

The obvious answer: Drive. Because the CAR needs to be at the car wash.

I tested the top models:

✅ Perplexity — nailed it instantly

✅ Grok — got it right

❌ Claude — said walk

❌ ChatGPT 5.2 — said walk (with a full essay about emissions and "European energy" 😭)

What's wild is that ChatGPT 5.2 is supposed to be smarter than ever. But I've noticed a pattern lately — it feels like it's gotten more confident and less logical at the same time.

Anyone else feel like ChatGPT has gotten noticeably dumber recently?

I think we're seeing the "sycophancy trap" in real time — models optimizing to sound helpful and fun rather than actually think through basic logic.

The car wash test isn't hard. It's a one-sentence reasoning check. If a model can't pass this, what does that say about how we're training these things?

Drop your results below 👇 Would love to know which model YOU think has regressed the most.