This Guide is about grinding stats in Final Fantasy Tactics by Level Up/Down.

There's some rumor where you de-level a character, and level it back up, and then through some mystery ??? you're supposed to be stronger and break the game. This guide aims to clarify about that with a boring wall of text and numbers and formulas.

Data is sourced from the wonderful “Battle Mechanics Guide by AeroStar” (BMG for short) from 20 years ago. That is the quintessential knowledge base that this guide will reference often. I can confirm that the game mechanics are mostly the same pertaining to stat growth. While that guide details every aspect of the battle system, this guide will have a narrower focus and expansion on stats.

It is my first time writing a guide, so if there's anything improper. Let me know as gently as possible. I'm fragile.

It is comparatively easy to grind a character to level 99. However, that only provides you the stat gains of 98 levels. FFT is a game of numbers and math. A few points of PA/MA have a big impact on damage. SP drives the turn order. Level Up/Down is the way to raise those numbers beyond their conventional ceiling. What else is there to do after maxing all jobs, have all items, but you still feel like grinding? Autism.

It’s already too easy to over-level in this game. Dorter might have caught people off guard, and Wiegraf is scary. However, people have done low level runs, people have done single class runs. Those honestly sound like a more enjoyable gameplay experience than stepping on and off the same tile. Plus, the whole process of grinding eats time like a Korean MMO. Do you have 70 hours for nothing? If you really want max stats, ffs just cheat engine or save edit. Tedium.

To briefly parrot the BMG, there’s 5 stats you can manipulate: HP, MP, SP, PA, MA. Those are displayed on your character screen and change depending on the primary job. The visible stats are a product an internal number (Raw) and a job multiplier (M). Excerpt from BMG 7.1

HP = [(RawHP * ClassHPMultiplier) / 1638400]
MP = [(RawMP * ClassMPMultiplier) / 1638400]
Sp = [(RawSp * ClassSpMultiplier) / 1638400]
PA = [(RawPA * ClassPAMultiplier) / 1638400]
MA = [(RawMA * ClassMAMultiplier) / 1683400]

Example: A level 99 Ramza Ninja:

Every time you level up or down, those internal Raw values are what’s changed. The Raw values drives the changes you see in your stat values. In the case of HP/MP, during battle; and in the case of SP/PA/MA, after battle. The formulas for stat gains and losses are as followed from the BMG 7.2, 7.3.

bonus = [current_RX / (C + Lv)]

decrement = [current_RX / (Lv + C)]

After some rephrasing and some high school math.


The core fact of the stat system is here. Traditional stat systems that adds +1, +2, +3 are very intuitive and familiar. In contrast, FFT uses a multiplicative, compounding system. Your stats grow by multiplying x1.01, x1.02, x1.03. The multipliers are also a chain of numerator and denominator pairs that cancel very nicely when multiple together in a sequence of level ups. Ignoring rounding, the overall stat change formulas leveling up from 1-99 simplify very nicely. The same is true for the change formula leveling down from 99-1.

This also makes quantifying gains difficult. I can't just say you'll gain 4 speed when you level up down. Gains and Losses are both proportional to your current stat. Things are in multipliers.

If C is very high to infinite, the stat ratio is 1. As such, the stats won’t change much no matter how much you level up or down. Conversely, if C is very low to 0, the 99 / 1 ratio takes over, yielding a very high multiplier. This is the reason for selecting classes with low C values for leveling up, and high C values for leveling down. The idea is to gain a lot going up, lose a little going down.

At equal C values within up/down cycles, you actually lose a bit of stats by the asymmetrical (– 1) quirk of the level down formula. Median is 6-8% for HP/MP; 0.5-1.5% for SP/PA/MA (for every complete cycle of 99-1-99).

Similarly, if level range is very high, the ratio will also be brought closer to 1. Thus the old observation of "level up/down works better on low levels" is true. If you limit yourself to 49 levels, 50-1-50 provides more gains than 99-50-99. Still, the data from this guide is geared towards doing the full range 99-1-99.

This is because, in practice it takes time and effort to stabilized for leveling up or down. Once it's done, you might as well go all the way. In cases where a level up portion is cut short unexpectedly, consider moving on to level down because the gains are already captured at the low levels. Conversely, always level down to level 1 because the ratios are most pronounced at the lowest levels.

Lastly, Leveling Down when you're already level 1 is likely the worst thing to do. Those stats are lost without any potential gain to counter-balance.

This is the basic table of C values for Generic classes.
Data is compiled from the BMG 7.4

This is a table of Stat change leveling up 1-99 or leveling down 99-1 levels.
Calculated from formulas in previous section.

All 5 stats grow and shrink together as a package deal. While job selection for leveling up and down is overwhelming. There are clear limitations to guide the way.

Thief and Ninja are the only generic classes with a low C value for SP. Most other classes have 100. In other words, the way to boost speed by leveling up as Ninjas. 1-99 will multiply your speed by 2.210. Leveling down from 99-1 at any other class is a divisor of 1.980. Divide Up multiplier by Down divisor is 2.210 / 1.980 is 1.116. Meaning each cycle of level down as not-Ninja and level up as Ninja gain you +11.6% more speed from when you started.

Similarly, the only way to raise MA is by leveling up as Mimes. 3.390 / 2.960 = 1.145 = +14.5% MA a cycle. Unfortunately, Mime has a very high relative C value for MP, which is bad for leveling. You tend to lose 30% of your MP every time you raise your MA. Unless the character has a unique class with a better spread (mostly just Reis), this is unavoidable for gaining MA. To offset loss in MP, go up Summoner or Wizard.

HP and PA come naturally from leveling down Bard and Dancer. Bard and Dancer has abysmal stat growth across the board. Those are the choice for level down class. I have no idea why people consider Chemist, but they come up in discussions.

The compounding nature of the Up/Down gains is why people aren’t even sure if it works. The majority of the gains happen at the end of the process. Rest assured that the game is breaking.

This is the basic table of M values for Generic classes.
Data is compiled from the BMG 7.4

255, 9999, 65535 are all familiar stat caps. Likewise, this game has stat caps. Internally, Raw stats are capped at 24 bits. 2^24-1 = 16,777,215. Externally, HP/MP are capped at 999, SP at 50, PA/MA at 99. Combined with the above M table, and the 7.1 stat formulas, functionally, your Raw stats are capped as follows:


With the above Raw spread, you would reach the maximum Stat at the worst multipliers from any job. Further gains will not have any effect. After all, Raw values are internal. Note that even at maximum Raw HP/MP, many classes don't have high enough multiplier to bring them to 999.

Those worst multipliers are Bard/Dancer for HP/MP; Arithmetician for SP; Bard for male PA; Arithmetician for female PA; Dragoon for MA. Why would you suffer just to give a lancer 99 MA? *shrug* This guide is concerned about observable maximums. Honestly, stop anytime.

One last bit of math. With the Raw stat cap being the ultimate goal. We’d have to know our starting point. You can kind of reverse estimate your stats across all classes and see where your Raw range is. There are details in the BMG about initial stats, but there’s variables and RNG and who remembers how they’ve leveled their Ramza. I haven’t found those useful, but maybe you have better comprehension.

Roughly, you want 8.5x SP and 13.5x MA. Mime loses a bit more MP than Ninja can gain, so you need some Summoner or Wizard ups at the end. HP and PA will cap themselves before SP starts to move. I’ve fully maxed Agrias once and Ramza once. The recipe ended up the same for both. Dancer works the same as Bard despite the difference in PA gain.

Expected stat changes from a complete 99-1-99 cycle

Estimated necessary cycles to perfect stats

Note that you can’t do all your Ninja levels at once. Gaining MP when capped wastes those gains. If your level 1 Bard/Dancer is over 63 MP, the 8x gain from 99 levels of Ninja will cap. So, you’d want to inject a Mime cycle then.

Lastly, an exception must be mentioned. Reis has an amazing stat gain spread. Dancer/Dragonner gains everything. Her recipe is the easiest by far.
Dragonner gains MA faster than Mimes, and even gains a bit of SP to reduce the Ninja cycles.

We need a place to level up, and a place to level down. For leveling up, pretty much anywhere works. For leveling down, you’d want a Degenerator trap. Messing around with Level Blast is possible. However, even to me, that sounds tedious. My choice is the Zeklaus Desert, my home for the past week. It opens after 4 maps in Chapter 1. In addition to the Degenerator trap, it also boasts a Sleeping Gas and a Steel Needle trap. It’s mostly flat, without too many dangerous enemies.

Walk on the Degenerator to lose a level. Walk off the Degenerator to reset. I've found it faster than messing about with failed Teleports. A helper unit can take a nap on Sleeping Gas by the cactus.

So that’s a lot about Level Up/Down. I thought the math behind it all was the coolest thing ever 20 years ago. I haven’t encountered this kind of leveling system in the 20 years since. If you want to try it out, one or two Bard/Ninja cycles is already plenty. You would have functionally level 150 stats in a game where enemies are barely level 50. Equipment won’t keep up and your fists will do more damage.