Is the new STM32G5 worth waiting for or should I just start with an F4?

The short answer is that if your primary goal is simply spinning a motor using a familiar workflow, stick with the F405 for now. The risk of spending time fighting the toolchain—the debugger setup and compiler quirks—on a brand new chip outweighs the benefits of the improved peripherals unless you specifically need the advanced features of the G5. The F4 series has a mature ecosystem where every obscure error has been solved and documented by someone else. You gain development velocity immediately.

Switching to the G5 means you're essentially running a beta version of an industrial-grade microcontroller. While the M33 core and improved timer structure are defintely better on paper for complex sensorless control algorithms, the real question is about the software stability on day one. ST is making huge strides, but moving to a brand new family usually involves initial stability issues in the HAL and the CubeMX setup that might not be documented yet. It takes time for the community and ST support channels to catch up and release the necessary errata fixes.

For a learning project, you should prioritize minimizing friction. If you are happy with the performance envelope of the F405, use it. If you are dead set on the G5's advanced capabilities, you need to budget significant time for tooling integration and validation. Download the latest software development kit, run a simple 'blink' project, then move to a basic timer setup. See if your existing probe and build environment work immediately. If they do, proceed, but assume you will be the first to discover some minor bugs. So, if you want guaranteed forward motion right now, go F405. If you want to test the bleeding edge, go G5, but expect a steeper learning curve and some inevitable frustrations while setting things up.

I had a similar moment last year when I picked a newer chip family over the older one everyone uses, and honestly the boards themselves weren't the problem, it was that half the tutorials online didn't apply so I spent ages reading the official manuals instead. The timers on the new one did seem nicer for motor stuff once I got there though. Do the example programs for the new chip include a motor demo or is that stuff still missing at launch?

The thing that actually got me moving on a newer ST family was buying one of thier cheap dev boards before committing to anything. I'd been reading specs for weeks and going nowhere, but once I had the board on my desk I just flashed their motor demo and poked at it in CubeIDE over a weekend. Ten bucks told me more than a month of forum lurking. If you order one of those for the new family and a similar board for the F part you're eyeing, you can test your exact workflow on both before deciding. That way you're not guessing whether your probe plays nice, you're checking it.

No advice here really, just chiming in because I'm in almost the same spot but with a stepper project instead of a motor with sensors. I kept waiting for the newer family to feel 'settled' and ended up stalling for months doing nothing at all. If I coud redo it I'd probaly just pick whichever chip has the example closest to what I'm building and start there. Does anyone know if the community libraries for the older chip handle closed loop stuff okay, or do most people end up rewriting that part anyway?