AI Desk · AI Fingerprinting · Compare
Google: Lyria 3 Clip Preview · Sakana: Fugu Max
Insufficient evidenceNot comparable yetAt least one of these records is declared unmeasurable (Sakana: Fugu Max: unmeasurable:caching), so no similarity verdict is published; the declared contracts below can still be compared.
Google: Lyria 3 Clip Preview recordSakana: Fugu Max recordOpen in the interactive bench
Tokenizer fingerprint
1 of 25 mutually clean strings agree| Test string | Google: Lyria 3 Clip Preview | Sakana: Fugu Max | Verdict |
|---|---|---|---|
| en-prose | 17 | 53 | differs |
| en-long | 24 | 62 | differs |
| spaces-20 | 7 | 9 | differs |
| spaces-60 | 17 | x | – |
| tabs-20 | 12 | 10 | differs |
| newlines-20 | 12 | 12 | match |
| mixed-ws | 5 | x | – |
| digits-9 | 3 | 17 | differs |
| digits-12 | 4 | x | – |
| digits-30 | 9 | x | – |
| digits-sep | 4 | x | – |
| float-long | x | x | – |
| zh-common | 19 | 37 | differs |
| zh-long | 25 | 46 | differs |
| zh-rare | 10 | x | – |
| ja-kana | 17 | 33 | differs |
| ja-kanji | 15 | 44 | differs |
| ko | 18 | 49 | differs |
| ru | 18 | 46 | differs |
| ar | 33 | 41 | differs |
| he | 25 | x | – |
| hi | 27 | 61 | differs |
| th | 27 | 53 | differs |
| el | 17 | 54 | differs |
| emoji-basic | 11 | 25 | differs |
| emoji-skin | 21 | 44 | differs |
| emoji-zwj-family | 12 | 33 | differs |
| emoji-zwj-x3 | 34 | x | – |
| emoji-flags | 17 | x | – |
| emoji-prof | 20 | x | – |
| math | 17 | x | – |
| boxdraw | 18 | x | – |
| combining | 7 | x | – |
| cjk-ext-b | 9 | x | – |
| surrogates | 28 | x | – |
| zalgo | 18 | x | – |
| rtl-mix | 13 | 21 | differs |
| py-code | 17 | x | – |
| py-indent | 21 | 60 | differs |
| json | 16 | x | – |
| html | 12 | x | – |
| regex | 17 | x | – |
| camel | 9 | 21 | differs |
| snake | 11 | 28 | differs |
| rare-word-x5 | 37 | 111 | differs |
| repeat-tok | 31 | 104 | differs |
| base64 | 13 | x | – |
| hex | 9 | x | – |
| url | 15 | x | – |
| uuid | 10 | x | – |
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: Lyria 3 Clip Preview | Sakana: Fugu Max | Verdict |
|---|---|---|---|
| Context length | 1048576 | 1000000 | × differs |
| Max output tokens | 65536 | 128000 | × differs |
| Supported parameters | max_tokens, response_format, seed, temperature, top_p | include_reasoning, reasoning, reasoning_effort, structured_outputs, tool_choice, tools, web_search_options | × differs |
| Defaults | {"temperature":null,"top_p":null,"top_k":null,"frequency_penalty":null,"presence_penalty":null,"repetition_penalty":null} | {"temperature":null,"top_p":null,"top_k":null,"frequency_penalty":null,"presence_penalty":null,"repetition_penalty":null} | ✓ match |
| Reasoning contract | – | {"mandatory":true,"default_enabled":true,"supported_efforts":["max","xhigh","high"],"default_effort":"high"} | × differs |
| Modalities | text+image->text+audio | text+image+file->text | × differs |
Method: tokenizer fingerprinting, v1, 50 strings. This page is static and dated; the interactive bench compares any two of the 337 measured models.