Bit depth: 24bit every time!
Sampling rate: If I'm aiming for audio only, I'll be using 88.2kHz. If aiming for video it's 96kHz. Just cause my lectures in mathematics and signal theory have taught me to favour an integral devider. (Not very complex knowledge, though.) I still deem 44.1kHz (and CD-Audio) as a standard for consumer-files.
EDIT: I don't think final products will have to be any higher than 48kHz.
Oversampling has helped me a lot, back in the day of DirectX and VST(1) Plugins - and so I tend to stick with it now. It's not the extra high frequency information that's been recorded (in fact I'm quite a fan of lowpass far below the nyquist). It's the output quality of digital audio processing and effects (in particular: aliasing), especially concerning heavy dynamics and analog modeling.
EDIT: So even if the final product is 44.1kHz, it made absolute sense to mix in 88.2 and then convert the final master.
Maybe I will rethink someday. Because this topic is also very entangled with the quality of the algorithms. (I tend to messure and know my plugs.) Moreover, in these days of high CPU power, a lot of the critical plugins allready use internal oversampling. So I'd probably be happy to use 44.1 on a lot of occasions.
I'll be receiving my new converters in February - and I'm planning on making some contemporary A/B tests, finally. (All steps from 44.1 up to 192.)