You asked Sonnet 5 to write a professional email. A person takes about 8 min. Sonnet 5 took 3.5 s, 136× faster, but "3.5 s" isn't one thing happening. Here's what actually filled that time, from a real recorded run, not a simulation.
First, the model reads what you asked. This isn't skimming, every word of your request gets turned into numbers it can work with, before it thinks about any of them.
Next, it works out what you're actually asking for: what kind of reply fits, what tone, what to include and what to leave out. This is the part that doesn't show up as any visible text.
Finally, it writes. This is the part you actually watch happen: the reply appearing on screen, word by word, at the speed you see in the recording.
The first word appeared 2.1 s in. Everything before that was reading and thinking, with nothing on screen yet.
Total: 3.5 s, versus 8 min for a person, on this one real run. Watch it yourself →
What it actually wrote
**Subject: Q3 Report – Feedback Needed by Friday** Hi Team, Please find the Q3 report attached for your review. Take a look when you get a chance, and send me your feedback, comments, or questions by end of day Friday. Your input is valuable in making sure we present accurate and polished results, so please don't hesitate to flag anything that needs clarification or correction. If Friday doesn't work for you, let me know as soon as possible so we can adjust the timeline accordingly. Thanks in advance for your time and input. Best regards, [Your Name]
One real example, not an average. How every stage is measured → · What is AI Speed? →