I used the Fluke 45 at work over 30 years ago. Recently I bought one to use on audio work (mainly analog tape decks) because it has better AC voltage specs as compared to the Fluke 289 or 189 that I have. It seems to have good accuracy up to 20kHz. To my surpise the input impedance for AC measurement is only 1 Megaohms and not 10 Megaohms like most other meter.
It's kind of a disappointment.

Actually that is pretty standard for bench meters.  1M DC-coupled across ~100pF.  Your 189 and 289 have 10M across ~100pF but with an AC blocking cap so infinite DC resistance.  Are the AC specs actually that much better?

189 and 289 are all 100khz frequency response

You could make up a 9Meg trimmable interface for a 10:1 divider with 10Meg input, and see what frequency response variation from specs you get using the other 2 flukes for comparison/calibration.

Actually that is pretty standard for bench meters.  1M DC-coupled across ~100pF.  Your 189 and 289 have 10M across ~100pF but with an AC blocking cap so infinite DC resistance.  Are the AC specs actually that much better?

In the audio range the AC specs is significantly better while the DC specs isn't.

In the audio range the AC specs is significantly better while the DC specs isn't.

Hmmm.  That 5V range, especially @ 10kHz-20kHz, is a bit odd for the 289, but other than that, once you factor in the counts and the 15% floor of the 45, I'm not sure your total error is all that much worse for the 289.

For example if you look a the tolerances for 2V @ 10kHz, the 289 is 0.6% + 25 counts on the 5V range, the 45 is 0.2% + 100 counts on the 10V range.  That's 14.5mV and 14mV respectively, a narrow win for the 45.

Now look at 250mV @ 1kHz, which gives you 0.4% + 25 vs 0.2% + 100.  That is 1.25mV vs 1.5mV, a win for the 289.

I don't think it would be valid to compare these DMMs without some calibration data.

I don't think it would be valid to compare these DMMs without some calibration data.

Why? Obviously if you need to prove the accuracy of your results, you need calibration traceability.  However, it is still valid to compare the specs.  For example, suppose you were trying to decide which of these meters to send off for calibration?  Then you'd want to know which of them would give better results, the assumption being that calibration would assure performance to specification by design, no better and hopefully no worse. 

What can be a bit misleading is that typically you'll see much better results than specified on the actual calibration tests.  That doesn't mean you can count on that level of performance under field conditions.

Here's a cal report for the AC ranges of a 289.

---

There are two ways to quantity uncertainty: Either you trust the manufacturer, and you take the specs as given, since they are guaranteed over a certain time, temperature, etc. Or you do your own characterization where you measure the performance over time, temperature, etc.

A single calibration report isn't sufficient for this. And it seems like an awful lot of effort for little gain. It might be better to spend the extra money on a unit with better specifications, then.

Obviously I'm not saying you can't compare these DMMs all day long, but without a reference or at least calibration data you're not going to be able to just blindly state that one is better than the other based on the published specifications.

I have various DMMs with current calibrations and some with data and have found specs to be a very questionable metric for comparisons, a topic that I still am amazed does not get more attention. You do have the occasional comment about certain  models, which reading further indicates that their popularity is due to them exceeding their specifications by a good margin.

Does the OP have any calibrated equipment or references in order to test the Fluke 45 vs the 289 or 189?  I would be interested to see the results.

You do have the occasional comment about certain  models, which reading further indicates that their popularity is due to them exceeding their specifications by a good margin.

True, but there are two important points that keep getting lost that I will continue to hammer on:

1.  When tested at the specified calibration temperature range, typically 23C +/- 1C, a meter absolutely needs to exceed its specifications by a significant margin to ensure that it will be in spec over the full specified temperature range and with the stated confidence interval for the specified calibration interval.  So if a meter has a "1-year" spec that adds up to say 15 counts at some specified calibration point with a 99% confidence interval, statistically the typical meter will be off less than 5 counts--assuming a standard distribution and yada yada.  There can be other factors involved in a stated tolerance spec as well, but that would be a whole book. 

2.  You cannot assume that just because the meter is very close--or even exact--at the calibration points and at the calibration temperature, that it is somehow guaranteed to exceed its published specification at some other voltage or some other temperature.  You would have to do a very extensive characterization of your own to make that claim.  In cases where that is done, it is typically done at some single test point for a particular purpose (DMM) or is an instrument with only one function (10V reference, standard resistor, etc). 

A calibration certificate for one or the other of those meters indicating that it was exactly correct at all of its test points (the best possible result) actually doesn't prove much, especially in light of the OP's stated purpose.