Grok 4.7 got chattier. Is that a good thing?

One of the loudest gripes about Grok 4.6 was its attitude. Short, dry answers, like it could not wait for you to leave. So what does 4.7 do? It talks. A lot. Independent testers at Artificial Analysis described it as notably slow and very verbose. Mission accomplished, right?

Well. Here is where it gets funny.

The pendulum swung

A new wave of complaints arrived, just pointing the other way. Developers testing it in their coding tools say it now rambles, buries the actual finding under a long intro, and assumes you know things it only saw in its own tool calls. One tester summed it up as picking up the bad habits of chattier rival models: overly wordy, less precise. What one person calls "finally, it actually talks to me", the next person calls "just get to the point".

Fine, it talks more. Who cares, if the price stayed the same?

Your wallet should. The price per token did stay the same as 4.6. But a chattier model produces more tokens for the same job. More tokens, same rate, bigger bill. One analysis put the real cost per task at roughly double. And it is slower too, around 57 tokens per second against 66 for the previous version. So you wait longer and pay more, for the same answer wrapped in more words.

This is the part of model design nobody puts on a launch slide. Every tweak to personality is a trade-off. Turn up the warmth and you turn up the word count, the latency and the cost. Turn it down and people say the model is cold. xAI fixed one complaint and walked straight into the opposite one. That is not a failure exactly. It is just what tuning a model's voice actually looks like.