That's been my experience with the BME688 too. Bosch likes to pitch this as being able to detect many various things, then leaves it up to the customer to experiment with heater profiles and machine learning to reach said goal.
The results I've seen for methane is discouraging, to say the least. Now moving to a NDIR sensor. A completely different thing, I know. Guess I was under a spell, it sounded too good to be true - a BOM cost reduction of 20x, and it was. I can go on and on about this one and the driverlib for it, but I've spent enough energy already.
SGP41 algorithm is open source, BME is closed source, requires you to link a binary during compilation.
In theory BME can sample additional gas profiles (programmable heater waveform), in practice it's useless.
I agree that the SGP41 has better sensitivity, and also faster response and better stability.