GPT-5.5 has officially dropped, and first impressions are promising.

Based on a quick 30-minute hands-on test, it feels like a genuine step forward with it being noticeably faster and noticeably smarter than its predecessor.

The numbers back that up. GPT-5.5 reportedly scores 4–7% higher than GPT-5.4 across multiple software benchmarks.

The broader AI coding community seems to agree on two things: the model is both faster and sharper in practice, not just on paper.

All is not sunshine and roses though, GPT-5.5 comes in at 2 TIMES the cost of GPT-5.4. A real upgrade, but one that comes with a real cost.

Whether it's the right tool depends entirely on how much that performance delta matters to you.

SpaceX has bought the option to acquire Cursor outright later this year for $60 billion, or pay $10 billion for the partnership if they pass on the buyout.

Cursor will train coding models on xAI’s GPUs and then likely share them with xAI and SpaceX.

For Cursor users, the obvious question is what this means for the product. Cursor has built a genuinely loyal following by being one of the best AI coding tools out there.

Getting folded into an Elon Musk-adjacent portfolio, alongside xAI and Grok, is not exactly what most of that user base signed up for.

Nothing is confirmed yet, and these things fall through all the time. But if this does happen, expect a messy conversation about data privacy, model training, and whether Cursor stays the product people actually love.

$60 billion is a lot to pay for a coding assistant. Then again, whoever controls the tools developers use every day controls quite a lot.

In the past day I have seen these headlines:

“Opus 4.7 is so dumb”

“Claude Opus 4.7 feels weird”

“Opus 4.7 is legendarily bad…”

Why is it that people think this?

The likely reason is a new feature called Adaptive Thinking, which tries to judge how hard the model needs to think for a given prompt.

The issue is it almost always lands on "not very hard" which is exactly the opposite of what people are paying Opus prices for.

The fix, for now, is a very non technical. You have to manually tell it to think properly. Something like "reason through this step by step" in your prompt usually does the trick. Not ideal, but it works.