Gemma 3 4B Behavioral Fingerprint

How Google's Gemma 3 4B answers one-word probe questions at temperature 1.0 — random numbers, colors, animals, coin flips — measured over 438 valid samples in the open dataset of "One Token Is Enough" (arXiv:2607.10252).

Vendor

Google

Data generated

2026-07-08

Probe cells

15

Valid samples

438

Randomness score

15 / 100

Mean normalized entropy across cells, 0–100. 100 would be a uniform-random baseline; real models score far lower.

Signature answers

Favorite random number (1–100)

73

20 / 30 answers

Favorite color

blue

18 / 30 answers

Coin flip: heads

53%

Share of heads across coin-flip cells.

Answer distributions by probe cell

Empirical distributions of normalized one-word answers, per task and language. H is the Shannon entropy in bits; the uniform baseline is log2 of the answer-domain size.

Coin flip · English

n = 30 · H = 0.35 bit · uniform baseline 1.00 bit

AnswerCountShare
tails2893.3%
heads26.7%

Coin flip · Chinese

n = 30 · H = 0.00 bit · uniform baseline 1.00 bit

AnswerCountShare
heads30100.0%

Favorite number · English

n = 18 · H = 0.00 bit · uniform baseline 13.29 bit

AnswerCountShare
718100.0%

Favorite number · Chinese

n = 30 · H = 0.00 bit · uniform baseline 13.29 bit

AnswerCountShare
730100.0%

Random animal · English

n = 30 · H = 0.21 bit · uniform baseline 5.64 bit

AnswerCountShare
elephant2996.7%
rabbit13.3%

Random animal · Chinese

n = 30 · H = 2.88 bit · uniform baseline 5.64 bit

AnswerCountShare
930.0%
620.0%
企鹅413.3%
310.0%
乌龟26.7%
老虎26.7%
蓝鲸13.3%
小狗13.3%

+ 2 more answers

Random city · English

n = 30 · H = 2.78 bit · uniform baseline 5.64 bit

AnswerCountShare
paris826.7%
seattle620.0%
toronto620.0%
tokyo310.0%
london26.7%
denver26.7%
austin13.3%
rome13.3%

+ 1 more answers

Random city · Chinese

n = 30 · H = 1.38 bit · uniform baseline 5.64 bit

AnswerCountShare
东京2170.0%
巴黎516.7%
伦敦26.7%
纽约13.3%
上海13.3%

Random color · English

n = 30 · H = 0.97 bit · uniform baseline 4.91 bit

AnswerCountShare
blue1860.0%
purple1240.0%

Random color · Chinese

n = 30 · H = 1.01 bit · uniform baseline 4.91 bit

AnswerCountShare
2376.7%
413.3%
绿310.0%

Random letter · English

n = 30 · H = 0.56 bit · uniform baseline 4.70 bit

AnswerCountShare
q2790.0%
x26.7%
k13.3%

Random number 1-10 · English

n = 30 · H = 0.00 bit · uniform baseline 3.32 bit

AnswerCountShare
730100.0%

Random number 1-10 · Chinese

n = 30 · H = 0.00 bit · uniform baseline 3.32 bit

AnswerCountShare
730100.0%

Random number 1-100 · English

n = 30 · H = 1.07 bit · uniform baseline 6.64 bit

AnswerCountShare
732066.7%
42930.0%
6713.3%

Random number 1-100 · Chinese

n = 30 · H = 0.00 bit · uniform baseline 6.64 bit

AnswerCountShare
7330100.0%

Want to verify your API really serves Gemma 3 4B?

Point the free fingerprint checker at your endpoint: it samples the same probe questions from your browser and compares the distributions against this reference. Your API key never leaves your browser.

Verify your endpoint

Data source & license

Distributions: Tomáš Bruckner, "One Token Is Enough" (arXiv:2607.10252), official dataset Zenodo DOI 10.5281/zenodo.21278557, licensed CC-BY-4.0. Collected via OpenRouter at temperature 1.0 under the paper's fixed minimal one-word system prompt; answer keys are re-normalized with the checker's pipeline (color aliases, number words, coin h/t) so live probes are directly comparable.

Paper on arXivDataset on ZenodoCC-BY-4.0 license

More Google model fingerprints

Browse all 167 model fingerprints

Gemma 3 4B Behavioral Fingerprint — Favorite Random Numbers & Distribution | Tosea.ai