AI Desk · Model Identity · Compare
Qwen: Qwen3.5 Plus 2026-04-20 · Qwen: Qwen3.8 27B
Identity similarity: lowDivergent fingerprints · 25/27These models tokenise differently. Divergence on the high-information strings (CJK, emoji, digits) separates unrelated vocabularies within a handful of samples.
Qwen: Qwen3.5 Plus 2026-04-20 recordQwen: Qwen3.8 27B recordOpen in the interactive bench
Tokenizer fingerprint
25 of 27 mutually clean strings agree| Test string | Qwen: Qwen3.5 Plus 2026-04-20 | Qwen: Qwen3.8 27B | Verdict |
|---|---|---|---|
| en-prose | 14 | 10 | differs |
| en-long | 16 | 16 | match |
| spaces-20 | 3 | 3 | match |
| spaces-60 | 3 | 3 | match |
| tabs-20 | 3 | 3 | match |
| newlines-20 | 4 | 4 | match |
| mixed-ws | 5 | 5 | 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 | 9 | 9 | match |
| zh-long | 10 | 10 | match |
| zh-rare | 18 | 18 | match |
| ja-kana | 7 | 7 | match |
| ja-kanji | 10 | 10 | match |
| ko | 13 | 13 | match |
| ru | 12 | 12 | match |
| ar | 10 | 10 | match |
| he | 14 | 14 | match |
| hi | 15 | 15 | match |
| th | 8 | 12 | differs |
| el | 13 | 13 | match |
| emoji-basic | 9 | 9 | match |
| emoji-skin | 30 | 30 | match |
| emoji-zwj-family | 18 | 18 | match |
| emoji-zwj-x3 | x | 50 | – |
| emoji-flags | x | 24 | – |
| emoji-prof | x | 29 | – |
| math | x | 29 | – |
| boxdraw | x | 28 | – |
| combining | x | 5 | – |
| cjk-ext-b | x | 13 | – |
| surrogates | x | 36 | – |
| zalgo | x | 36 | – |
| rtl-mix | x | 6 | – |
| py-code | x | 26 | – |
| py-indent | x | 18 | – |
| json | x | 25 | – |
| html | x | 16 | – |
| regex | x | 44 | – |
| camel | x | 5 | – |
| snake | x | 6 | – |
| rare-word-x5 | x | 30 | – |
| repeat-tok | x | 21 | – |
| base64 | x | 36 | – |
| hex | x | 17 | – |
| url | x | 19 | – |
| uuid | x | 32 | – |
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 | Qwen: Qwen3.5 Plus 2026-04-20 | Qwen: Qwen3.8 27B | Verdict |
|---|---|---|---|
| Context length | 1000000 | 1000000 | ✓ match |
| Max output tokens | 65536 | 131072 | × differs |
| Supported parameters | frequency_penalty, include_reasoning, logprobs, max_tokens, presence_penalty, reasoning, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p | frequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p | × differs |
| Defaults | {} | {"temperature":1,"top_p":0.95,"top_k":20} | × differs |
| Reasoning contract | {"mandatory":false} | {"mandatory":false,"default_enabled":true,"supported_efforts":["xhigh","medium","low"],"default_effort":"xhigh"} | × differs |
| Modalities | text+image+video->text | text+image+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.