I recently tested both Grok-3 and DeepSeek with what seemed like a simple task:

Write a 50-word story using ONLY words starting with 'S'.

Default mode

Both AIs failed their first attempts:

Grok-3: 47 words (4 wrong ones)

DeepSeek: 53 words (7 wrong ones)

I tried 3 more times, but they kept messing up. 

DeepSeek attempts:

Grok 3 attempts:

Then I turned on "think mode" and things got interesting.

Think mode

DeepSeek was super fast (12 seconds) but still couldn't get it right. 49 words and 1 of them wrong.

Grok-3 took its sweet time (1min 40s), but here's the cool part - it started fixing its previous mistakes and finally nailed it with this perfect story:

"Smart, sophisticated system starts searching. System sifts signals systematically. Signals show significant sequences. Such sequences suggest solutions. Solutions solve situations successfully. System succeeds splendidly. Scientists see system's superior skills. System's success stuns scientists significantly. Scientists study system's structure secretly. System's secrets seem sacred. System stays silent, seeming satisfied, showing strength."

Exactly 50 words, all starting with 'S'. Pretty neat how Grok-3 took longer but got it perfect while DeepSeek was quick but couldn't quite get there.

TL;DR: Sometimes slower but more thorough wins the race. Grok-3 might take longer, but it seems to be better at complex language tasks that require precise rules to be followed.