Buying skills by buying modes is much more efficient than buying them with improvement points.
In the full-on inject-the-math-directly-into-my-veins character creation mode of Atomic Robo, you have 30 improvement points to spend at character creation. Buying a mode at any level costs its value in improvement points, even though your Good (+3) mode has three times the rating of your Average (+1) mode.
So, just as an example, suppose these three modes: Omnipotent (20) which contains every skill in the game, Cheeseball (2) which only contains Combat, and Son of Cheeseball (2), a completely different mode which also only contains Combat.
It costs the same number of character creation points to be +3 Omnipotent +2 Cheeseball +1 Son of Cheeseball as it does to be +3 Son of Cheeseball +2 Cheeseball +1 Omnipotent, and you'll have 6 improvement points left over. Both characters will be at +5 Combat, but the +3 Omnipotent character will have every other skill in the game at +3, while the +1 Omnipotent character will only have them at +1.
Or for a less egregious example, look at Jenkins (p. 256) and work out what his final skill totals would be if he was +3 Survivor +2 Action +1 Jenkins instead of the other way round.
If you deliberately break character creation, character creation will be deliberately broken.
Of course, if you show up to a game with +3 Omnipotent +2 Cheeseball +1 Son of Cheeseball, the GM will tremble in helpless terror at the correctness of your mathematics laugh you out of the building, but it's still worth considering that a big disparity in mode costs is a big disparity in character power right out the gate.
This is why the game tries to cap most of its modes at 9 or 10 points' worth of skills. Aside from Jenkins, which, c'mon, it's Jenkins. Some of the prebaked weird modes in game are pretty low in points and high in improvements - Pilot and Robot, for example, and there's even one, Mutant, which is made up entirely of improvements. You will have a less powerful character if you take one of those modes at Good or Fair rather than at Average, especially if you duplicate all of their skills in a different mode.
Some modes try to make up for a low skill count by having a unique skill in them that you can't get any other way, but just as a general rule, when you're constructing a weird mode, try and make a mode that people will "expect" to spend 8 or 9 points in. Science, for example, is only a 3-point mode, but you can see how it might "expect" you to spend 9 points total by getting two specializations in Science, or one specialization and three focuses. Something like Psychokinetic, with a 2-point weird skill and a 2-point standard skill, is a little underpointed even assuming people might want to specialize in Psychokinetic.
How much does this actually matter?
Probably very little, unless you take Good (+3) Mutant because your character concept is that you're a mutant first, or build an entirely weird-mode character and don't get enough skills to actually spend all your improvements on. Anything else will at least see you out of character creation with a half dozen skills at +3 or better, which is all you really need to assert yourself in a game of Atomic Robo.
Only For Robo
It's also worth noting that this is not how the reference example of modes actually works. The imbalance here is largely because Robo modes aren't balanced against each other from a skill access perspective, while Fate modes should be if you're doing them stock. There isn't the option in the reference example to effectively give up a mode selection for more improvement points; you get a fixed number and it's assumed modes are balanced around that number.
The advice still stands that if you're designing your own Fate modes and some of them drop skill selections for extras, you should still make sure every mode has a reason to be picked first.
