AI Desk · Model Identity · Compare
Anthropic: Claude 3 Haiku · Reka Edge
Identity similarity: lowDivergent fingerprints · 12/50These models tokenise differently. Divergence on the high-information strings (CJK, emoji, digits) separates unrelated vocabularies within a handful of samples.
Anthropic: Claude 3 Haiku recordReka Edge recordOpen in the interactive bench
Tokenizer fingerprint
12 of 50 mutually clean strings agree| Test string | Anthropic: Claude 3 Haiku | Reka Edge | Verdict |
|---|---|---|---|
| en-prose | 15 | 14 | differs |
| en-long | 17 | 15 | differs |
| spaces-20 | 4 | 3 | differs |
| spaces-60 | 6 | 3 | differs |
| tabs-20 | 4 | 3 | differs |
| newlines-20 | 3 | 4 | differs |
| mixed-ws | 9 | 5 | differs |
| digits-9 | 4 | 4 | match |
| digits-12 | 5 | 5 | match |
| digits-30 | 11 | 11 | match |
| digits-sep | 8 | 8 | match |
| float-long | 10 | 10 | match |
| zh-common | 19 | 19 | match |
| zh-long | 25 | 26 | differs |
| zh-rare | 21 | 24 | differs |
| ja-kana | 12 | 15 | differs |
| ja-kanji | 17 | 17 | match |
| ko | 20 | 19 | differs |
| ru | 14 | 16 | differs |
| ar | 21 | 25 | differs |
| he | 18 | 24 | differs |
| hi | 23 | 32 | differs |
| th | 25 | 26 | differs |
| el | 22 | 28 | differs |
| emoji-basic | 11 | 9 | differs |
| emoji-skin | 26 | 29 | differs |
| emoji-zwj-family | 15 | 17 | differs |
| emoji-zwj-x3 | 43 | 53 | differs |
| emoji-flags | 25 | 24 | differs |
| emoji-prof | 28 | 28 | match |
| math | 47 | 31 | differs |
| boxdraw | 29 | 28 | differs |
| combining | 6 | 14 | differs |
| cjk-ext-b | 17 | 14 | differs |
| surrogates | 55 | 40 | differs |
| zalgo | 36 | 36 | match |
| rtl-mix | 10 | 11 | differs |
| py-code | 30 | 25 | differs |
| py-indent | 19 | 15 | differs |
| json | 23 | 20 | differs |
| html | 17 | 17 | match |
| regex | 52 | 44 | differs |
| camel | 7 | 5 | differs |
| snake | 11 | 6 | differs |
| rare-word-x5 | 41 | 31 | differs |
| repeat-tok | 21 | 21 | match |
| base64 | 43 | 35 | differs |
| hex | 14 | 11 | differs |
| url | 22 | 17 | differs |
| uuid | 28 | 28 | 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 | Anthropic: Claude 3 Haiku | Reka Edge | Verdict |
|---|---|---|---|
| Context length | 200000 | 16384 | × differs |
| Max output tokens | 4096 | 16384 | × differs |
| Supported parameters | max_tokens, stop, temperature, tool_choice, tools, top_k, top_p | frequency_penalty, logprobs, max_tokens, presence_penalty, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, 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} | × differs |
| Modalities | text+image->text | text+image+video->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.