AI Desk · Model Identity · Compare
Amazon: Nova 2 Lite · Meta: Llama 4 Scout
Identity similarity: lowDivergent fingerprints · 30/50These models tokenise differently. Divergence on the high-information strings (CJK, emoji, digits) separates unrelated vocabularies within a handful of samples.
Amazon: Nova 2 Lite recordMeta: Llama 4 Scout recordOpen in the interactive bench
Tokenizer fingerprint
30 of 50 mutually clean strings agree| Test string | Amazon: Nova 2 Lite | Meta: Llama 4 Scout | Verdict |
|---|---|---|---|
| en-prose | 14 | 14 | match |
| en-long | 15 | 17 | differs |
| spaces-20 | 3 | 3 | match |
| spaces-60 | 3 | 3 | match |
| tabs-20 | 3 | 3 | match |
| newlines-20 | 4 | 4 | match |
| mixed-ws | 6 | 6 | match |
| digits-9 | 3 | 3 | match |
| digits-12 | 4 | 4 | match |
| digits-30 | 10 | 10 | match |
| digits-sep | 7 | 7 | match |
| float-long | 9 | 9 | match |
| zh-common | 12 | 11 | differs |
| zh-long | 16 | 12 | differs |
| zh-rare | 18 | 18 | match |
| ja-kana | 8 | 9 | differs |
| ja-kanji | 10 | 12 | differs |
| ko | 12 | 10 | differs |
| ru | 11 | 10 | differs |
| ar | 10 | 11 | differs |
| he | 10 | 10 | match |
| hi | 11 | 12 | differs |
| th | 9 | 8 | differs |
| el | 14 | 14 | match |
| emoji-basic | 10 | 10 | match |
| emoji-skin | 11 | 20 | differs |
| emoji-zwj-family | 11 | 11 | match |
| emoji-zwj-x3 | 33 | 33 | match |
| emoji-flags | 14 | 16 | differs |
| emoji-prof | 18 | 19 | differs |
| math | 25 | 24 | differs |
| boxdraw | 24 | 21 | differs |
| combining | 10 | 10 | match |
| cjk-ext-b | 16 | 16 | match |
| surrogates | 32 | 37 | differs |
| zalgo | 34 | 31 | differs |
| rtl-mix | 6 | 6 | match |
| py-code | 25 | 25 | match |
| py-indent | 15 | 15 | match |
| json | 20 | 20 | match |
| html | 16 | 16 | match |
| regex | 42 | 42 | match |
| camel | 6 | 6 | match |
| snake | 6 | 6 | match |
| rare-word-x5 | 30 | 30 | match |
| repeat-tok | 21 | 40 | differs |
| base64 | 31 | 35 | differs |
| hex | 10 | 11 | differs |
| url | 17 | 17 | match |
| uuid | 27 | 27 | 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 | Amazon: Nova 2 Lite | Meta: Llama 4 Scout | Verdict |
|---|---|---|---|
| Context length | 1000000 | 1310720 | × differs |
| Max output tokens | 65535 | 16384 | × differs |
| Supported parameters | include_reasoning, max_tokens, reasoning, stop, temperature, tool_choice, tools, top_k, top_p | frequency_penalty, logit_bias, max_tokens, min_p, presence_penalty, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_p | × differs |
| Defaults | {"temperature":null,"top_p":null,"frequency_penalty":null} | {} | × differs |
| Reasoning contract | {"mandatory":false} | – | × differs |
| Modalities | text+image+file+video->text | text+image->text | × differs |
Method: tokenizer fingerprinting, v1, 50 strings. Raw data on the data page. This page is static and dated; the interactive bench compares any two of the 422 catalogue models.