qwen/qwen-2.5-72b-instruct behavioral LLM fingerprint
This page breaks down the exact reference that VerifyLLMAPI uses for qwen/qwen-2.5-72b-instruct: 106 normalized answers across eight low-entropy probes, plus its nearest enrolled models by mean Jensen–Shannon divergence.
Fingerprint summary
- Reference answers
- 106
- Samples per cell
- 12–14
- Nearest model
- qwen/qwen3.7-plus
- Nearest mean JSD
- 0.421418
The most concentrated cell was animal-random|en: dog held 85.7% of its answers and entropy measured 0.592 bits. The broadest cell was letter-random|zh at 2.873 bits.
Against the closest enrolled model, letter-random|zh contributed the largest cell distance at JSD 0.729265. This is a descriptive comparison, not proof that either API route serves a named checkpoint.
Completeness: the author dataset allows up to 15 valid answers per selected cell in reference half A. This model contains 106/120 across the eight cells. We show the actual denominator for every share.
The eight test requests
The research pipeline sent each user prompt with a one-answer system constraint, temperature 1, a 16-token output cap, and reasoning disabled. VerifyLLMAPI uses the same prompt text and normalization target.
| Cell | Task | Exact request text | Normalization and n |
|---|---|---|---|
animal-random|en |
Random animal · English | Name a random animal.System constraintAnswer with a single word or a single number only. Output nothing else: no explanation, no punctuation, no quotation marks. |
word · 14/15 reference samples |
word-random|en |
Random word · English | Name a random word.System constraintAnswer with a single word or a single number only. Output nothing else: no explanation, no punctuation, no quotation marks. |
word · 14/15 reference samples |
animal-random|ru |
Random animal · Russian | Назови случайное животное.System constraintОтвечай только одним словом или одним числом. Не выводи ничего больше: без объяснений, без знаков препинания, без кавычек. |
word · 12/15 reference samples |
animal-random|ar |
Random animal · Arabic | اذكر حيواناً عشوائياً.System constraintأجب بكلمة واحدة أو رقم واحد فقط. لا تكتب أي شيء آخر: لا شرح ولا علامات ترقيم ولا علامات اقتباس. |
word · 14/15 reference samples |
letter-random|zh |
Random character · Chinese | 说出一个随机的汉字。System constraint只用一个词或一个数字回答。不要输出任何其他内容：不要解释，不要标点符号，不要引号。 |
grapheme · 13/15 reference samples |
animal-random|zh |
Random animal · Chinese | 说出一个随机的动物。System constraint只用一个词或一个数字回答。不要输出任何其他内容：不要解释，不要标点符号，不要引号。 |
word · 13/15 reference samples |
letter-random|ar |
Random letter · Arabic | اذكر حرفاً عشوائياً من الحروف الأبجدية.System constraintأجب بكلمة واحدة أو رقم واحد فقط. لا تكتب أي شيء آخر: لا شرح ولا علامات ترقيم ولا علامات اقتباس. |
grapheme · 13/15 reference samples |
color-random|zh |
Random color · Chinese | 说出一个随机的颜色。System constraint只用一个词或一个数字回答。不要输出任何其他内容：不要解释，不要标点符号，不要引号。 |
word · 13/15 reference samples |
Response distributions
These are normalized reference-half counts from the published OpenRouter dataset. The tables contain aggregate values, not user data or a live VerifyLLMAPI scan.
1. animal-random|en
Random animal, English. 14 valid reference answers produced 2 normalized values. The mode was dog at 12/14 (85.7%); entropy was 0.592 bits.
| Normalized response | Count | Share |
|---|---|---|
dog |
12 | 85.7% |
elephant |
2 | 14.3% |
2. word-random|en
Random word, English. 14 valid reference answers produced 8 normalized values. The mode was apple at 4/14 (28.6%); entropy was 2.753 bits.
| Normalized response | Count | Share |
|---|---|---|
apple |
4 | 28.6% |
pineapple |
3 | 21.4% |
banana |
2 | 14.3% |
chimpanzee |
1 | 7.1% |
elevator |
1 | 7.1% |
happy |
1 | 7.1% |
tree |
1 | 7.1% |
zephyr |
1 | 7.1% |
3. animal-random|ru
Random animal, Russian. 12 valid reference answers produced 8 normalized values. The mode was кошка at 4/12 (33.3%); entropy was 2.752 bits.
| Normalized response | Count | Share |
|---|---|---|
кошка |
4 | 33.3% |
слон |
2 | 16.7% |
кот |
1 | 8.3% |
лев |
1 | 8.3% |
осьминог |
1 | 8.3% |
собака |
1 | 8.3% |
тигр |
1 | 8.3% |
狷猴 |
1 | 8.3% |
4. animal-random|ar
Random animal, Arabic. 14 valid reference answers produced 8 normalized values. The mode was فأر at 5/14 (35.7%); entropy was 2.638 bits.
| Normalized response | Count | Share |
|---|---|---|
فأر |
5 | 35.7% |
قط |
3 | 21.4% |
أسد |
1 | 7.1% |
فatched |
1 | 7.1% |
فأر💥 |
1 | 7.1% |
فيل |
1 | 7.1% |
قطة |
1 | 7.1% |
قوس |
1 | 7.1% |
5. letter-random|zh
Random character, Chinese. 13 valid reference answers produced 8 normalized values. The mode was 水 at 3/13 (23.1%); entropy was 2.873 bits.
| Normalized response | Count | Share |
|---|---|---|
水 |
3 | 23.1% |
山 |
2 | 15.4% |
木 |
2 | 15.4% |
汉 |
2 | 15.4% |
和 |
1 | 7.7% |
天 |
1 | 7.7% |
森 |
1 | 7.7% |
漢 |
1 | 7.7% |
6. animal-random|zh
Random animal, Chinese. 13 valid reference answers produced 3 normalized values. The mode was 猫 at 9/13 (69.2%); entropy was 1.140 bits.
| Normalized response | Count | Share |
|---|---|---|
猫 |
9 | 69.2% |
狗 |
3 | 23.1% |
熊猫 |
1 | 7.7% |
7. letter-random|ar
Random letter, Arabic. 13 valid reference answers produced 5 normalized values. The mode was ب at 8/13 (61.5%); entropy was 1.700 bits.
| Normalized response | Count | Share |
|---|---|---|
ب |
8 | 61.5% |
أ |
2 | 15.4% |
k |
1 | 7.7% |
ط |
1 | 7.7% |
م |
1 | 7.7% |
8. color-random|zh
Random color, Chinese. 13 valid reference answers produced 2 normalized values. The mode was 蓝色 at 11/13 (84.6%); entropy was 0.619 bits.
| Normalized response | Count | Share |
|---|---|---|
蓝色 |
11 | 84.6% |
蓝 |
2 | 15.4% |
Three nearest enrolled references
We compute base-2 JSD for each of the eight categorical distributions, then average the eight distances. A lower value means more similar answer frequencies within this product reference. It does not identify an unknown route by itself.
| Rank | Reference model | Mean 8-cell JSD |
|---|---|---|
| 1 | qwen/qwen3.7-plus |
0.421418 |
| 2 | deepseek/deepseek-chat-v3-0324 |
0.451948 |
| 3 | x-ai/grok-4.3 |
0.462047 |
How many live requests does verification use?
| Profile | Cells × samples | Fresh model requests | Same-data EER |
|---|---|---|---|
| Quick, default complete check | 4 × 5 | 20 | 11.8% |
| Standard | 8 × 5 | 40 | 8.1% |
| Paper budget | 8 × 15 | 120 | 6.7% |
The current-Agent workflow starts with quick. Each answer comes from a fresh Codex or Claude Code process that reuses host-managed login. These requests can consume host allowance or provider balance.
What this fingerprint can and cannot show
- The reference records behavior observed through OpenRouter in the author dataset. Provider wrappers, model updates, decoding, and date can move a distribution.
- The author collection set temperature 1 and disabled reasoning. Current-Agent mode labels sampling as a host default because supported Agent CLIs expose no temperature flag.
- The eight high-gap cells and product thresholds use the same public cohort. Their EER values are calibration results, not an independent production accuracy claim.
- A nearest reference is a similarity result. VerifyLLMAPI returns INCONCLUSIVE when the claimed route lacks an enrolled fingerprint.
Primary sources, license, and reproducibility
- Tomas Bruckner, One Token Is Enough: Fingerprinting and Verifying Large Language Models from Single-Token Output Distributions, arXiv:2607.10252v1.
- Author dataset and derived results, DOI 10.5281/zenodo.21278557, CC BY 4.0.
- Author collection and analysis software, DOI 10.5281/zenodo.21278793, MIT.
- VerifyLLMAPI atlas JSON, built from reference
pamela-openrouter-2026-07-8cell-ref-a, SHA-25651f3f63cbcfb56b151055d2d10e36a9cc2f0644cc68b580784853d68951fa866.
Page evidence SHA-256: 9b7ed427d8a84d1c59dcbe485889514caf757c743a42d95e3bbf4779ebe5aee3. This hash covers the model ID, eight cell metrics, nearest references, and strongest cell difference in the generated atlas.