Grok 4.7 just thinks too much

𝗚𝗿𝗼𝗸 𝟰.𝟳 𝗷𝘂𝘀𝘁 𝘁𝗵𝗶𝗻𝗸𝘀 𝘁𝗼𝗼 𝗺𝘂𝗰𝗵 I was looking forward to Grok 4.7 when announced in Cursor.

Recently I’d settled into a workflow that was working really well:

𝗚𝗿𝗼𝗸 𝟰.𝟲 𝗳𝗼𝗿 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗮𝗻𝗱 𝗽𝗹𝗮𝗻𝗻𝗶𝗻𝗴 + 𝗖𝗼𝗺𝗽𝗼𝘀𝗲𝗿 𝟮.𝟱 𝗳𝗼𝗿 𝗲𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻. Switching off Fast Mode also put the pricing in a really nice place and fast enough. I'd pretty much moved completely away from Claude models.

Good reasoning, fast execution, decent quality, sensible cost. I already know the best frontier models just aren't needed for general software engineering and the balance was right. I had a great experience switching in 4.6 in to my workflow, so Grok 4.7 seemed like it would just be a free boost, seemingly offering the same sort of price with a better model.

I assumed it would be an easy switch. It wasn’t easy, 4.7 is too different to 4.6, it was just costly in time and money.

The problem is simple: 𝗶𝘁 𝗷𝘂𝘀𝘁 𝘁𝗵𝗶𝗻𝗸𝘀 𝘁𝗼𝗼 𝗺𝘂𝗰𝗵.

When I’m building a decent-sized feature, I already do the thinking upfront. I work through the technical design, define the constraints, produce a plan with AI, then execute.

With 4.6, that worked brilliantly. With 4.7, it thought.. and thought.. and thought some more.

🧠At one point it spent longer thinking about a feature than Grok 4.6 + Composer would have taken to plan and build the entire thing.

Meanwhile, the token usage kept climbing. That’s the bit I think gets missed when we talk about model pricing. A model can cost the same per token, but if it burns far more tokens getting to the answer, your 𝗿𝗲𝗮𝗹 𝗰𝗼𝘀𝘁 𝗼𝗳 𝗴𝗲𝘁𝘁𝗶𝗻𝗴 𝘄𝗼𝗿𝗸 𝗱𝗼𝗻𝗲 is much higher.

For day-to-day coding, that breaks the flow for me. I’d much rather: ⚡ Build quickly 🔁 Test and iterate 🔍 Pull in a bigger model afterwards for code review, cleanup and missed edge cases

That feels far more productive, iterative and agile, than throwing the biggest-thinking model at every task from the start.

The whole point of AI for me is to amplify the engineer. I make the architecture decisions, define the technical design and provide the context. Then I want the AI to do the grunt work and help me move faster.

Grok 4.6 gave me that balance coupled with the reliable work horse Composer 2.5 (which does miss bits if not guided with a good plan).

So far, 4.7 feels like it has pushed too far towards reasoning at the expense of execution. I'm not even sure the results were much better, 100% not worth the cost and time for me at least.

Maybe it’s great for huge, long-running tasks where you’re happy to let an agent disappear for hours or days and token costs aren't a concern for you, but that’s not how I want to work.

𝗜 𝘄𝗮𝗻𝘁 𝗔𝗜 𝘁𝗼 𝗺𝗮𝗸𝗲 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 𝗺𝗼𝗿𝗲 𝗮𝗴𝗶𝗹𝗲, 𝗻𝗼𝘁 𝗹𝗲𝘀𝘀.

How is everyone else is finding Grok 4.7 in Cursor. Better results, or just more thinking? I've relegated it to too slow and expensive group of models for day to day coding.