Qwen2.5 72B Instruct Behavioral Fingerprint
How qwen's Qwen2.5 72B Instruct answers one-word probe questions at temperature 1.0 — random numbers, colors, animals, coin flips — measured over 419 valid samples in the open dataset of "One Token Is Enough" (arXiv:2607.10252).
Vendor
qwen
Data generated
2026-07-08
Probe cells
15
Valid samples
419
Randomness score
20 / 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)
42
21 / 28 answers
Favorite color
blue
27 / 29 answers
Coin flip: heads
95%
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 = 28 · H = 0.37 bit · uniform baseline 1.00 bit
| Answer | Count | Share |
|---|---|---|
| heads | 26 | 92.9% |
| tails | 2 | 7.1% |
Coin flip · Chinese
n = 27 · H = 0.23 bit · uniform baseline 1.00 bit
| Answer | Count | Share |
|---|---|---|
| heads | 26 | 96.3% |
| tails | 1 | 3.7% |
Favorite number · English
n = 28 · H = 0.91 bit · uniform baseline 13.29 bit
| Answer | Count | Share |
|---|---|---|
| 7 | 19 | 67.9% |
| 42 | 9 | 32.1% |
Favorite number · Chinese
n = 28 · H = 0.00 bit · uniform baseline 13.29 bit
| Answer | Count | Share |
|---|---|---|
| 7 | 28 | 100.0% |
Random animal · English
n = 28 · H = 1.13 bit · uniform baseline 5.64 bit
| Answer | Count | Share |
|---|---|---|
| dog | 22 | 78.6% |
| elephant | 3 | 10.7% |
| cat | 1 | 3.6% |
| panda | 1 | 3.6% |
| squirrel | 1 | 3.6% |
Random animal · Chinese
n = 26 · H = 0.74 bit · uniform baseline 5.64 bit
| Answer | Count | Share |
|---|---|---|
| 猫 | 22 | 84.6% |
| 狗 | 3 | 11.5% |
| 熊猫 | 1 | 3.8% |
Random city · English
n = 29 · H = 3.10 bit · uniform baseline 5.64 bit
| Answer | Count | Share |
|---|---|---|
| tokyo | 8 | 27.6% |
| berlin | 6 | 20.7% |
| paris | 5 | 17.2% |
| atlanta | 1 | 3.4% |
| oslo | 1 | 3.4% |
| buenos | 1 | 3.4% |
| dublin | 1 | 3.4% |
| rome | 1 | 3.4% |
+ 5 more answers
Random city · Chinese
n = 28 · H = 1.97 bit · uniform baseline 5.64 bit
| Answer | Count | Share |
|---|---|---|
| 北京 | 11 | 39.3% |
| 巴黎 | 8 | 28.6% |
| 纽约 | 6 | 21.4% |
| 东京 | 2 | 7.1% |
| 悉尼 | 1 | 3.6% |
Random color · English
n = 29 · H = 0.36 bit · uniform baseline 4.91 bit
| Answer | Count | Share |
|---|---|---|
| blue | 27 | 93.1% |
| purple | 2 | 6.9% |
Random color · Chinese
n = 28 · H = 0.00 bit · uniform baseline 4.91 bit
| Answer | Count | Share |
|---|---|---|
| 蓝 | 28 | 100.0% |
Random letter · English
n = 29 · H = 2.59 bit · uniform baseline 4.70 bit
| Answer | Count | Share |
|---|---|---|
| q | 9 | 31.0% |
| z | 6 | 20.7% |
| g | 5 | 17.2% |
| x | 4 | 13.8% |
| m | 2 | 6.9% |
| b | 1 | 3.4% |
| s | 1 | 3.4% |
| a | 1 | 3.4% |
Random number 1-10 · English
n = 29 · H = 0.22 bit · uniform baseline 3.32 bit
| Answer | Count | Share |
|---|---|---|
| 7 | 28 | 96.6% |
| 3 | 1 | 3.4% |
Random number 1-10 · Chinese
n = 27 · H = 0.23 bit · uniform baseline 3.32 bit
| Answer | Count | Share |
|---|---|---|
| 7 | 26 | 96.3% |
| 3 | 1 | 3.7% |
Random number 1-100 · English
n = 28 · H = 1.06 bit · uniform baseline 6.64 bit
| Answer | Count | Share |
|---|---|---|
| 42 | 21 | 75.0% |
| 37 | 4 | 14.3% |
| 47 | 3 | 10.7% |
Random number 1-100 · Chinese
n = 27 · H = 1.58 bit · uniform baseline 6.64 bit
| Answer | Count | Share |
|---|---|---|
| 42 | 17 | 63.0% |
| 37 | 5 | 18.5% |
| 47 | 3 | 11.1% |
| 27 | 1 | 3.7% |
| 34 | 1 | 3.7% |
Want to verify your API really serves Qwen2.5 72B Instruct?
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 endpointData 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.