probably mods i prefer 1.7.10 for mods though but i think 1.5.2 is better if you want no lag your not losing a lot of features only main thing is horses really they aren't really great you still get hungry while riding anyways 1.7.10 and 1.5.2 are both good versions i didn't know 1.6.4 was the best though
1.6.4 lags a log less, and some prefer the older biome generation style, which includes the notorious 'infinite oceans'. For mods, I prefer it over 1.7.10 in terms of performance, but doesn't have much choice, and its mods don't get updated often.
I mainly prefer it because of the underground generation; cave systems were much larger and denser compared to newer versions, as well as more varied (you could even find areas with less caves than any comparable area in newer versions, as seen in this comparison). While the larger cave systems may seem very large I can explore them within a couple play sessions, covering an average of about 100 chunks per play session. Other underground features were also made rarer (there are about 40% as many mineshafts and dungeons, slightly offset in 1.13, which removed a reduction in mineshaft frequency within 1280 blocks of the origin, but this only makes them more common within 512 blocks).
Mojang also could have implemented the new biomes a lot better; with the way that I play I'd hardly see much, and less variety (meaning very different biome types, e.g. desert and snow) unless I chose a seed; for example, here is a comparison between a modded world and the same seed in 1.7:
I don't care much about oceans (and since 1.13 they are completely off-limits since they no longer have any real caves), and if anything I've reduced them in my own mods, but they are still a true global ocean (I mainly see an issue with seeds that spawn you in an ocean or with little land nearby; they did try to avoid this in vanilla but the algorithm that "grows" land "seeds" doesn't always generate a landmass around the origin; I changed it so there is guaranteed to be land within 1000 blocks of it).
Of course, as others mentioned the performance is another factor, especially given how newer versions ran on my older computer, but there is still a big difference; for example, I can run 1.6.4 at 150 FPS on 20 chunk render distance with only 512 MB of memory allocated, and server tick time is only about 4 ms, 8% of the maximum (this may be biased by the fact this this includes many of my own optimizations, which more than offset added features); the game also starts up and generates worlds much faster (despite 1.13 having multithreaded world generation, which may really only mean that a single dedicated thread is being used, not multiple threads):
This also includes mods; given the requirements that I see for modern modded versions, such as 4+ GB of RAM just to start, there is absolutely no way I could ever hope to run them; Mojang has also recently increased the default allocation for vanilla to 2 GB (for 64 bit), and in either case this makes no sense (loaded chunks account for about 200 MB of the memory usage in the example above). I blame Forge as well as Mojang for this; my modus operandi is that any new features added must not impact performance, or as little as possible, and the much simpler code in older versions is much easier to optimize (most of my changes, some giving an order of magnitude improvement, make little real changes to the original code):
The old Notch code was straightforward and relatively easy to follow. The new rendering system is an over-engineered monster full of factories, builders, bakeries, baked items, managers, dispatchers, states, enums and layers. Object allocation is rampant, small objects are allocated like there is no tomorrow. No wonder that the garbage collector has to work so hard.
1.6.4 does have an issue with zombie pathfinding causing lag (even a few zombies unable to reach a target causes very noticeable server lag) but I fixed it years ago, before Mojang declared it fully fixed, and either way it is situational. Non-Nvidia GPUs may also have issues with the "Advanced OpenGL" setting, which enables GPU-based occlusion culling (I blame Intel and AMD though, which often still have issues due to their poor OpenGL support. 1.8 and later use a CPU-based approach which is supposed to work better but it is also buggier).
As far as content goes, I don't care for most of the features that have been added since, or even available in 1.6.4, as these statistics for a "caving only" world show (in some of my worlds I just started caving without starting out from scratch; either way, most of what I do is directly related to caving); even all the features I've added to TMCW were mostly added for fun, or to show how Mojang could have done something (for example, they way they added new blocks which caused them to run out of block IDs much faster; by contrast, I've added several hundred new blocks, most as variants of existing blocks so I've only used a couple dozen actual block IDs with over 60 left; or making Mending work as a direct replacement for renaming an item to keep the cost down, meaning you have to use resources to repair items as well as be smart about how many enchantments you add; and emphasizing mining instead of farming for resources).
Likewise, while newer versions may have fixed some bugs I've fixed many of the more noticeable ones myself (often earlier, using fixes posted to the bug tracker), without all the new bugs that have been added since then (IIRC 1.13 was said to be responsible for like 1/3 of all bugs ever reported. That's not to say that my own code is bug-free; possibly the most significant bug in TMCW was one that caused a small percentage of worlds to not have any mineshafts more than 640 blocks away from the origin, due to a bug with the code that checked for strongholds, which I only found when somebody showed me a stronghold and mineshaft overlapping, which wasn't supposed to happen. I wouldn't say that is as bad as the vanilla code though, which has issues affecting virtually all seeds, which appears to be worse in 1.13, as seen in the seed for my first world (every mineshaft shown appears to match its sign-reversed counterpart in 1.13 as they all have the same absolute coordinates when mirrored across an axis, while only one (verified from an actual world, the coordinates can match without the same structure but that isn't likely for the number in 1.13) matches in 1.6).