YFarmX

AI Desk · Model Identity · Compare

Google: Gemini 2.5 Flash · Google: Gemini 3.7 Flash

Identity similarity: lowDivergent fingerprints · 26/26These models tokenise differently. Divergence on the high-information strings (CJK, emoji, digits) separates unrelated vocabularies within a handful of samples.

Google: Gemini 2.5 Flash recordGoogle: Gemini 3.7 Flash recordOpen in the interactive bench

Tokenizer fingerprint

26 of 26 mutually clean strings agree
Test stringGoogle: Gemini 2.5 FlashGoogle: Gemini 3.7 FlashVerdict
en-prose1414match
en-long1616match
spaces-2033match
spaces-6044match
tabs-2033match
newlines-2033match
mixed-ws99match
digits-999match
digits-121212match
digits-303030match
digits-sep1212match
float-long2222match
zh-common1010match
zh-long1212match
zh-rare2525match
ja-kana77match
ja-kanji99match
ko1111match
ru1212match
ar1010match
he1313match
hi88match
th1010match
el1313match
emoji-basic55match
emoji-skin1010match
emoji-zwj-familyx7
emoji-zwj-x3x21
emoji-flagsx8
emoji-profx11
mathx26
boxdrawx17
combiningx10
cjk-ext-bx16
surrogatesx27
zalgox22
rtl-mixx5
py-codex28
py-indentx18
jsonx28
htmlx14
regexx44
camelx6
snakex11
rare-word-x5x25
repeat-tokx21
base64x34
hexx16
urlx23
uuidx36

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
FieldGoogle: Gemini 2.5 FlashGoogle: Gemini 3.7 FlashVerdict
Context length10485761048576✓ match
Max output tokens6553565536× differs
Supported parametersinclude_reasoning, max_tokens, reasoning, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_pinclude_reasoning, max_tokens, reasoning, reasoning_effort, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_p× differs
Defaults{"temperature":null,"top_p":null,"top_k":null,"frequency_penalty":null,"presence_penalty":null,"repetition_penalty":null}{}× differs
Reasoning contract{"mandatory":false}{"mandatory":true,"default_enabled":true,"supported_efforts":["high","medium","low"],"default_effort":"medium"}× differs
Modalitiestext+image+file+audio+video->texttext+image+file+audio+video->text✓ match

Method: tokenizer fingerprinting, v1, 50 strings. This page is static and dated; the interactive bench compares any two of the 453 catalogue models.