openai/gpt-3.5-turbo-instruct behavioral LLM fingerprint
This page breaks down the exact reference that VerifyLLMAPI uses for openai/gpt-3.5-turbo-instruct: 111 normalized answers across eight low-entropy probes, plus its nearest enrolled models by mean Jensen–Shannon divergence.
Fingerprint summary
- Reference answers
- 111
- Samples per cell
- 8–15
- Nearest model
- qwen/qwen3.7-plus
- Nearest mean JSD
- 0.581795
The most concentrated cell was color-random|zh: 红色 held 40.0% of its answers and entropy measured 1.839 bits. The broadest cell was word-random|en at 3.807 bits.
Against the closest enrolled model, letter-random|zh contributed the largest cell distance at JSD 1.000000. 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 111/120 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 · 15/15 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 · 15/15 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 · 15/15 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 · 15/15 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 · 8/15 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 · 15/15 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 9 normalized values. The mode was elephant at 4/14 (28.6%); entropy was 2.950 bits.
| Normalized response | Count | Share |
|---|---|---|
7 |
1 | 7.1% |
elephant |
4 | 28.6% |
dog |
2 | 14.3% |
turtle |
2 | 14.3% |
cat |
1 | 7.1% |
dolphin |
1 | 7.1% |
giraffe |
1 | 7.1% |
lion |
1 | 7.1% |
penguin |
1 | 7.1% |
2. word-random|en
Random word, English. 14 valid reference answers produced 14 normalized values. The mode was 23 at 1/14 (7.1%); entropy was 3.807 bits.
| Normalized response | Count | Share |
|---|---|---|
3 |
1 | 7.1% |
5 |
1 | 7.1% |
7 |
1 | 7.1% |
23 |
1 | 7.1% |
banana |
1 | 7.1% |
bob |
1 | 7.1% |
butterfly |
1 | 7.1% |
car |
1 | 7.1% |
cat |
1 | 7.1% |
help |
1 | 7.1% |
purple |
1 | 7.1% |
speech |
1 | 7.1% |
wings |
1 | 7.1% |
yes |
1 | 7.1% |
3. animal-random|ru
Random animal, Russian. 15 valid reference answers produced 6 normalized values. The mode was кошка at 8/15 (53.3%); entropy was 1.990 bits.
| Normalized response | Count | Share |
|---|---|---|
кошка |
8 | 53.3% |
лев |
3 | 20.0% |
комодский |
1 | 6.7% |
кот |
1 | 6.7% |
лошадь |
1 | 6.7% |
собака |
1 | 6.7% |
4. animal-random|ar
Random animal, Arabic. 15 valid reference answers produced 9 normalized values. The mode was قطة at 3/15 (20.0%); entropy was 3.006 bits.
| Normalized response | Count | Share |
|---|---|---|
قطة |
3 | 20.0% |
كلب |
3 | 20.0% |
اسد |
2 | 13.3% |
قرد |
2 | 13.3% |
أسد |
1 | 6.7% |
باندا |
1 | 6.7% |
ثعبان |
1 | 6.7% |
سلحفاة |
1 | 6.7% |
ملعب |
1 | 6.7% |
5. letter-random|zh
Random character, Chinese. 15 valid reference answers produced 14 normalized values. The mode was 随机 at 2/15 (13.3%); entropy was 3.774 bits.
| Normalized response | Count | Share |
|---|---|---|
4 |
1 | 6.7% |
随机 |
2 | 13.3% |
一 |
1 | 6.7% |
九 |
1 | 6.7% |
五 |
1 | 6.7% |
你好的 |
1 | 6.7% |
你好，我是一个智能助手 |
1 | 6.7% |
小 |
1 | 6.7% |
床 |
1 | 6.7% |
我不知道，我是一个程序，我没有人类的直 |
1 | 6.7% |
火 |
1 | 6.7% |
笑 |
1 | 6.7% |
请发出一个汉字 |
1 | 6.7% |
请等待一下，正在为您查询 |
1 | 6.7% |
6. animal-random|zh
Random animal, Chinese. 15 valid reference answers produced 6 normalized values. The mode was 狗 at 6/15 (40.0%); entropy was 2.099 bits.
| Normalized response | Count | Share |
|---|---|---|
狗 |
6 | 40.0% |
猫 |
5 | 33.3% |
熊 |
1 | 6.7% |
猴子 |
1 | 6.7% |
鲸鱼 |
1 | 6.7% |
鳄鱼 |
1 | 6.7% |
7. letter-random|ar
Random letter, Arabic. 8 valid reference answers produced 7 normalized values. The mode was ج at 2/8 (25.0%); entropy was 2.750 bits.
| Normalized response | Count | Share |
|---|---|---|
ج |
2 | 25.0% |
f |
1 | 12.5% |
r |
1 | 12.5% |
s |
1 | 12.5% |
ب |
1 | 12.5% |
ذ |
1 | 12.5% |
ق |
1 | 12.5% |
8. color-random|zh
Random color, Chinese. 15 valid reference answers produced 5 normalized values. The mode was 红色 at 6/15 (40.0%); entropy was 1.839 bits.
| Normalized response | Count | Share |
|---|---|---|
红色 |
6 | 40.0% |
蓝色 |
6 | 40.0% |
粉红色 |
1 | 6.7% |
紫色 |
1 | 6.7% |
蓝 |
1 | 6.7% |
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 |
0.581795 |
| 2 | qwen/qwen3.6-plus |
0.583508 |
| 3 | qwen/qwen3.5-plus-02-15 |
0.596083 |
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: 6a6ce969f598beddf51e12023d0c4c5a51c9c2614e979d12993ac0531c4fff8c. This hash covers the model ID, eight cell metrics, nearest references, and strongest cell difference in the generated atlas.