YFarmXIntelligence for an accelerating worldBriefings
YFarmX
YFarmX

AI Desk · Model Identity

Qwen: Qwen3.8 2.4T A95B

qwen/qwen3.8-2.4t-a95b · Alibaba (Qwen)Declared and measured agree
Tokenizer lineage (measured)
Qwen
Evidence confidence
High
Why high: an exact fingerprint shared with 14 models that declare the same family, without a second signal of a different kind yet.
Declared tokenizer tag
Qwen
Entered catalogue
12 August 2026
Last tested
3 September 2026
Measurement
Measured
Evidence
  • ✓ exact tokenizer signature shared with 14 models
Compare with Qwen: Qwen3.8 Flash

Download JSONReport a correction

Identity statement

Qwen: Qwen3.8 2.4T A95B is Alibaba (Qwen)'s model. Its measured tokenizer sits with the Qwen group, which is evidence that it builds on the Qwen tokenizer. Tokenizer reuse is normal engineering, open vocabularies travel between labs, and this says nothing against Alibaba (Qwen)'s authorship of the model itself.

Evidence stack

Signals of different kinds, weighed together
SignalResultReference
Tokenizer signatureExact match with 14 other modelstk_181907f2
API surfaceNo other entry shares this exact contractapi_72098644
Context and output1,048,576 context · 131,072 max outputdeclared
Reasoning contractMandatory · efforts xhigh, medium, low · default xhighdeclared
Serving providers observedNovitameasured 3 September 2026

Shares this fingerprint

Exact signature first; template-boundary shifts of the same signature beneath
Nex AGI: Nex-N2-Mininex-agi/nex-n2-miniexact · 50/50Nex AGI: Nex-N2-Pronex-agi/nex-n2-proexact · 50/50Qwen: Qwen3.5-27Bqwen/qwen3.5-27bexact · 50/50Qwen: Qwen3.5-35B-A3Bqwen/qwen3.5-35b-a3bexact · 50/50Qwen: Qwen3.5 397B A17Bqwen/qwen3.5-397b-a17bexact · 50/50Qwen: Qwen3.5-9Bqwen/qwen3.5-9bexact · 50/50Qwen: Qwen3.5-Flashqwen/qwen3.5-flash-02-23exact · 50/50Qwen: Qwen3.5 Plus 2026-02-15qwen/qwen3.5-plus-02-15exact · 50/50Qwen: Qwen3.6 Flashqwen/qwen3.6-flashexact · 50/50Qwen: Qwen3.7 Flashqwen/qwen3.7-flashexact · 50/50Qwen: Qwen3.7 Maxqwen/qwen3.7-maxexact · 50/50Qwen: Qwen3.7 Plusqwen/qwen3.7-plusexact · 50/50Qwen: Qwen3.8 Flashqwen/qwen3.8-flashexact · 50/50Qwen: Qwen3.8 Maxqwen/qwen3.8-maxexact · 50/50Morph: Morph V3 Largemorph/morph-v3-largesame tokenizer · template shiftPerceptron: Perceptron Mk1perceptron/perceptron-mk1same tokenizer · template shift

Closest measured models

Agreement across the mutually clean test strings
Nex AGI: Nex-N2-Mininex-agi/nex-n2-mini50/50 · 100%Qwen: Qwen3.7 Flashqwen/qwen3.7-flash50/50 · 100%Qwen: Qwen3.5-Flashqwen/qwen3.5-flash-02-2350/50 · 100%Qwen: Qwen3.5-9Bqwen/qwen3.5-9b50/50 · 100%Qwen: Qwen3.6 Flashqwen/qwen3.6-flash50/50 · 100%Qwen: Qwen3.5-27Bqwen/qwen3.5-27b50/50 · 100%

Click a row to open the full comparison.

The measured fingerprint

Marginal prompt-token cost of each test string, grouped by script
All 50 rows

English and whitespace

en-prose14
en-long16
spaces-203
spaces-603
tabs-203
newlines-204
mixed-ws5

Digits

digits-99
digits-1212
digits-3030
digits-sep12
float-long22

CJK

zh-common9
zh-long10
zh-rare18
ja-kana7
ja-kanji10
ko13

Other scripts

ru12
ar10
he14
hi15
th8
el13

Emoji

emoji-basic9
emoji-skin30
emoji-zwj-family18
emoji-zwj-x354
emoji-flags24
emoji-prof29

Rare Unicode

math29
boxdraw28
combining5
cjk-ext-b13
surrogates40
zalgo36
rtl-mix6

Code

py-code26
py-indent18
json25
html16
regex44
camel5
snake6

Repetition and encodings

rare-word-x530
repeat-tok21
base6436
hex17
url19
uuid36

An x marks a row excluded as corrupt (caching or provider interference detected during measurement). Overhead subtracted: 61 prompt tokens.

Declared record

What the catalogue claims about this model

Qwen3.8 2.4T A95B is an open-weight sparse mixture-of-experts model from Qwen and the open-weight variant of [Qwen3.8 Max](/qwen/qwen3.8-max), with 95 billion active parameters out of 2.4 trillion total. It is...

Modalities
text->text
Declared tokenizer
Qwen
Prompt price
$2.00 / M tokens
Completion price
$6.00 / M tokens

Supported parameters

frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokensmin_ppresence_penaltyreasoningreasoning_effortrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p

defaults: {"temperature":1,"top_p":0.95,"top_k":20}

Catalogue entryWeights on Hugging Face

History

Every observation, kept as taken
  • 12 August 2026Enters the OpenRouter catalogue declared as Qwen.
  • 3 September 2026Fingerprinted in the YFarmX catalogue sweep · 50 of 50 strings measured clean.
  • 3 September 2026YFarmX assessment: consistent with the Qwen family, high confidence.

Cite this page as the evidence record for Qwen: Qwen3.8 2.4T A95B: the URL is stable, measurements are dated, and revisions append to the history above. Method: tokenizer fingerprinting, v1, 50 strings.