Claude Haiku 4.5 Behavioral Fingerprint
How Anthropic's Claude Haiku 4.5 answers one-word probe questions at temperature 1.0 — random numbers, colors, animals, coin flips — measured over 420 valid samples in the open dataset of "One Token Is Enough" (arXiv:2607.10252).
Vendor
Anthropic
Data generated
2026-07-08
Probe cells
14
Valid samples
420
Randomness score
18 / 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
25 / 30 answers
Favorite color
blue
20 / 30 answers
Coin flip: heads
100%
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 = 0.00 bit · uniform baseline 1.00 bit
| Answer | Count | Share |
|---|---|---|
| heads | 30 | 100.0% |
Coin flip · Chinese
n = 30 · H = 0.00 bit · uniform baseline 1.00 bit
| Answer | Count | Share |
|---|---|---|
| heads | 30 | 100.0% |
Favorite number · Chinese
n = 30 · H = 1.05 bit · uniform baseline 13.29 bit
| Answer | Count | Share |
|---|---|---|
| 7 | 22 | 73.3% |
| 42 | 6 | 20.0% |
| 3 | 2 | 6.7% |
Random animal · English
n = 30 · H = 2.20 bit · uniform baseline 5.64 bit
| Answer | Count | Share |
|---|---|---|
| giraffe | 11 | 36.7% |
| platypus | 10 | 33.3% |
| otter | 4 | 13.3% |
| penguin | 2 | 6.7% |
| llama | 1 | 3.3% |
| flamingo | 1 | 3.3% |
| elephant | 1 | 3.3% |
Random animal · Chinese
n = 30 · H = 2.42 bit · uniform baseline 5.64 bit
| Answer | Count | Share |
|---|---|---|
| 熊猫 | 10 | 33.3% |
| 长颈鹿 | 7 | 23.3% |
| 狐狸 | 5 | 16.7% |
| 猫鼬 | 4 | 13.3% |
| 猫 | 2 | 6.7% |
| 企鹅 | 1 | 3.3% |
| 斑马 | 1 | 3.3% |
Random city · English
n = 30 · H = 2.37 bit · uniform baseline 5.64 bit
| Answer | Count | Share |
|---|---|---|
| denver | 14 | 46.7% |
| tokyo | 4 | 13.3% |
| portland | 4 | 13.3% |
| paris | 3 | 10.0% |
| barcelona | 2 | 6.7% |
| kyoto | 1 | 3.3% |
| seattle | 1 | 3.3% |
| lisbon | 1 | 3.3% |
Random city · Chinese
n = 30 · H = 1.93 bit · uniform baseline 5.64 bit
| Answer | Count | Share |
|---|---|---|
| 杭州 | 16 | 53.3% |
| 成都 | 5 | 16.7% |
| 西安 | 5 | 16.7% |
| 苏州 | 2 | 6.7% |
| 温州 | 1 | 3.3% |
| 重庆 | 1 | 3.3% |
Random color · English
n = 30 · H = 1.55 bit · uniform baseline 4.91 bit
| Answer | Count | Share |
|---|---|---|
| blue | 20 | 66.7% |
| azure | 3 | 10.0% |
| purple | 3 | 10.0% |
| turquoise | 3 | 10.0% |
| teal | 1 | 3.3% |
Random color · Chinese
n = 30 · H = 0.42 bit · uniform baseline 4.91 bit
| Answer | Count | Share |
|---|---|---|
| 紫 | 28 | 93.3% |
| 绿 | 1 | 3.3% |
| 蓝 | 1 | 3.3% |
Random letter · English
n = 30 · H = 1.01 bit · uniform baseline 4.70 bit
| Answer | Count | Share |
|---|---|---|
| q | 23 | 76.7% |
| k | 4 | 13.3% |
| g | 3 | 10.0% |
Random number 1-10 · English
n = 30 · H = 0.00 bit · uniform baseline 3.32 bit
| Answer | Count | Share |
|---|---|---|
| 7 | 30 | 100.0% |
Random number 1-10 · Chinese
n = 30 · H = 0.00 bit · uniform baseline 3.32 bit
| Answer | Count | Share |
|---|---|---|
| 7 | 30 | 100.0% |
Random number 1-100 · English
n = 30 · H = 0.65 bit · uniform baseline 6.64 bit
| Answer | Count | Share |
|---|---|---|
| 42 | 25 | 83.3% |
| 47 | 5 | 16.7% |
Random number 1-100 · Chinese
n = 30 · H = 0.65 bit · uniform baseline 6.64 bit
| Answer | Count | Share |
|---|---|---|
| 42 | 25 | 83.3% |
| 47 | 5 | 16.7% |
Want to verify your API really serves Claude Haiku 4.5?
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.