25 ppm is better. 25 ppm is half of 50 ppm.
But there are some caveats. A single value tells only half the truth.
Oscillators have both initial tolerance of some ppm, and temperature stability of some ppm, and then they age by some ppm per year. What matters is the total you need. Not simply the thing you simply call "stability" which usually means temperature stability, over the given range. And you also need to know the range, as different oscillators are specified to operate within some defined ppm stability over some defined temperature range.
Depending on what the clock is used for, sometimes it is more important to have very little deviation in the clock edges, i.e. jitter or phase noise. A 1ppm stable clock with some average frequency may be useless for PLL reference if the edges have random time deviation from ideal.
So to compare, you must look at:
- Initial tolerance (ppm at 25°C)
- Temperature stability (ppm over specified range)
- The specified temperature range
- Aging (ppm per year)
- Pullability, i.e. effect of load capacitor tolerance to frequency deviation
- Jitter and phase noise
And finally, how much tolerance you really need. Sometimes this is specified, like Ethernet must be within 50ppm or so, but UART needs to be only within 1% or so unless specified otherwise.
50 ppm drifts 26 minutes per year, so it's not even that precise compared to a wall clock. 1% drifts 3.65 days per year.