AI Desk · AI Fingerprinting · Compare
Google: Gemini 3.8 Flash · Google: Gemma 3n 4B
Identity similarity: highNear-exact agreement · 48/50Agreement this close usually means one tokenizer behind two serving templates, or a small vocabulary revision inside one family. The differing rows below say which strings part.
Google: Gemini 3.8 Flash recordGoogle: Gemma 3n 4B recordOpen in the interactive bench
Tokenizer fingerprint
48 of 50 mutually clean strings agree| Test string | Google: Gemini 3.8 Flash | Google: Gemma 3n 4B | Verdict |
|---|---|---|---|
| en-prose | 14 | 14 | match |
| en-long | 16 | 16 | match |
| spaces-20 | 3 | 3 | match |
| spaces-60 | 4 | 4 | match |
| tabs-20 | 3 | 3 | match |
| newlines-20 | 3 | 3 | match |
| mixed-ws | 9 | 9 | match |
| digits-9 | 9 | 9 | match |
| digits-12 | 12 | 12 | match |
| digits-30 | 30 | 30 | match |
| digits-sep | 12 | 12 | match |
| float-long | 22 | 22 | match |
| zh-common | 10 | 10 | match |
| zh-long | 12 | 12 | match |
| zh-rare | 25 | 25 | match |
| ja-kana | 7 | 7 | match |
| ja-kanji | 9 | 9 | match |
| ko | 11 | 11 | match |
| ru | 12 | 12 | match |
| ar | 10 | 10 | match |
| he | 13 | 13 | match |
| hi | 8 | 8 | match |
| th | 10 | 10 | match |
| el | 13 | 13 | match |
| emoji-basic | 5 | 5 | match |
| emoji-skin | 10 | 10 | match |
| emoji-zwj-family | 7 | 7 | match |
| emoji-zwj-x3 | 21 | 21 | match |
| emoji-flags | 8 | 8 | match |
| emoji-prof | 11 | 11 | match |
| math | 26 | 26 | match |
| boxdraw | 17 | 17 | match |
| combining | 10 | 10 | match |
| cjk-ext-b | 16 | 16 | match |
| surrogates | 27 | 27 | match |
| zalgo | 22 | 22 | match |
| rtl-mix | 5 | 5 | match |
| py-code | 28 | 28 | match |
| py-indent | 18 | 18 | match |
| json | 28 | 28 | match |
| html | 14 | 14 | match |
| regex | 44 | 44 | match |
| camel | 6 | 6 | match |
| snake | 11 | 11 | match |
| rare-word-x5 | 25 | 26 | differs |
| repeat-tok | 21 | 22 | differs |
| base64 | 34 | 34 | match |
| hex | 16 | 16 | match |
| url | 23 | 23 | match |
| uuid | 36 | 36 | match |
Values are the marginal prompt-token cost of each string with the model's fixed overhead subtracted. An x marks a row excluded as corrupt for that model.
API surface
The declared serving contracts differ| Field | Google: Gemini 3.8 Flash | Google: Gemma 3n 4B | Verdict |
|---|---|---|---|
| Context length | 1048576 | 32768 | × differs |
| Max output tokens | 65536 | – | × differs |
| Supported parameters | include_reasoning, max_tokens, reasoning, reasoning_effort, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_p | frequency_penalty, logit_bias, max_tokens, min_p, presence_penalty, repetition_penalty, response_format, stop, structured_outputs, temperature, top_k, top_p | × differs |
| Defaults | {} | {} | ✓ match |
| Reasoning contract | {"mandatory":true,"default_enabled":true,"supported_efforts":["high","medium","low"],"default_effort":"medium"} | – | × differs |
| Modalities | text+image+file+audio+video->text | text->text | × differs |
Method: tokenizer fingerprinting, v1, 50 strings. This page is static and dated; the interactive bench compares any two of the 338 measured models.