GachaGuide

Genshin Inferred Probabilities


Version: v3.15.1

Disclaimer

Genshin Impact publishes base rates, consolidated rates and hard pity limits per banner. This information alone is not enough to accurately model wishes. Our tooling is based on probabilities inferred from data obtained by community or our own team. These probabilities are our best guess at pity mechanics and are tuned to match the published consolidated rates. Our calculators should match the long-term behavior of actual in-game wishes. Sources are cited below. More detailed rationale can be found in our deep dive.

Sources

[1] Whale watching logs

[2] Non-uniform soft pity

[3] Linear ramping simulation

[4] GachaGuide study on four star pity

Banner data

[5] paimon-moe banner history (MIT)

[6] Genshin Impact Wiki, per-banner pages (CC BY-SA)

Standard and Character Banner


Five Star Probabilities per Pity, consolidated Rate: 1.60%

WishProbability
10.006
2 - 730.006
740.064
750.122
760.181
770.239
780.298
790.356
800.415
810.473
820.532
830.590
840.649
850.707
860.766
870.824
880.883
890.941
901.000

Uses linear ramping soft pity at 74th pull - per [3]. This results in a close match with the reported consolidated rate and fits the histogram in [1].

Four Star Probabilities per Pity, consolidated Rate: 13.0%

WishProbability
10.051
20.051
30.051
40.051
50.051
60.051
70.051
80.051
90.52
101.000

All pulls have 5.1% base rate until the 9th and 10th. The 10th has a 100% chance (hard pity) and the 9th pull probability was chosen to achieve a consolidated rate of 13.00%.

The five star 50/50

Since version 5.0 the 50/50 on character banners isn't quite a 50/50 any more. A five star won while you're not on a guarantee turns out to be the promotional character 55.000% of the time, and miHoYo publishes that figure along with the rule that you can never lose four in a row, but not the mechanic that produces either of them.

So we're modelling it as a 4.212% chance that a lost 50/50 gets converted into the promotional character, which becomes a guarantee once you've lost three in a row (it's the only constant trigger rate that gives us both published numbers at once). That works out to a promotional share of 68.966% of all five stars rather than 2/3, and 90.33 expected wishes per promotional character instead of 93.45, which is a decent chunk if you're budgeting.

One wrinkle and it's ours rather than the game's. The simulation is starting you at a loss streak of zero, so your first 50/50 will show up as 52.1% and not 55.000%. The 55.000% is what you converge to once streaks are in play. There's no input for your current streak yet.

Capturing Radiance is character banner only, so weapon and standard aren't affected at all.

Weapon Banner


Five Star Probabilities per Pity, consolidated rate: 1.85%

WishProbability
10.007
2 - 630.007
640.065
650.123
660.182
670.240
680.299
690.357
700.415
710.474
720.532
730.591
740.649
750.707
760.766
770.824
780.883
790.941
801.000

Uses linear ramping soft pity at 64th pull - per [3] but shifted. This results in a close match with the reported consolidated rate (1.85%) and fits the histogram in [1].

Four Star Probabilities per Pity, consolidated rate: 14.5%

WishProbability
10.06
20.06
30.06
40.06
50.06
60.06
70.06
80.41
91.000
10n/a

Uses tuned probabilities from our own study [4] to match histogram shape. Base probability of 6% alone with hard pity at 9th pull does not give a 14.5% rate. From our data there is additional soft pity on the 8th pull. That value (0.41) is chosen to achieve the published consolidated rate of 14.5%.

Epitomized Path

Charting a course guarantees the weapon you picked, and since version 5.0 it only takes one Fate Point instead of two. Any five star weapon that isn't your choice earns you the point, so your weapon shows up on the first five star weapon 37.5% of the time and on the second one the other 62.5%. Never later than that. Two five star weapons, 160 wishes worst case, which used to be 240.

Our calculator actually models this, you tick the box on a five star weapon to set your choice.

Yeah it appears that community tables show 56.1% on the 9th wish for the character banner and 66.0% on the 8th for the weapon banner (from the Fandom wiki and GGanalysis, out of a 9.45 million wish study) but we chose 0.52 and 0.41 because they reproduce really close to miHoYo's published rates. I know there is a decent delta but I'd rather make sure we converge to published rates in our model. Their numbers give consolidated rates of 13.06% and 14.86%, ours give 13.01% and 14.50%, and the published figures are 13.0% and 14.5%.

Where these numbers are used