Passer au contenu principal

Qwen3.6 35B A3B (Reasoning)

AlibabaQwenOpen WeightApache 2.0 · Usage Commercial

Description

Qwen3.6-35B-A3B is the first open-weight variant of the Qwen3.6 series, a multimodal Mixture-of-Experts model with 35B total parameters and 3B activated. It pairs a vision encoder with a hybrid 40-layer language model that interleaves Gated DeltaNet linear-attention blocks and Gated Attention blocks (10 × (3 × DeltaNet + 1 × Attention)) over 256 experts (8 routed + 1 shared, expert dim 512). The release prioritizes stability and real-world utility, with substantial gains in agentic coding (frontend workflows, repo-level reasoning) and a new option to preserve reasoning context across turns. Native context length is 262K tokens, extensible to ~1M via YaRN, and the model thinks by default.

Date de sortie
2026-04-16
Paramètres
35.0B
Longueur du contexte
262K
Modalités
image, text, video

Radar de capacités

70
general
50
coding
80
reasoning
77
scienceest.
50
agents
80
multimodal

Science est estimé à partir des scores scientifiques de LLM Stats ou du raisonnement lorsque les benchmarks scientifiques dédiés ne sont pas disponibles.

Classements

Domaine#RangScoreSource
Capacité agentique119
38.0
LS
Classement multimodal37
60.0
LS

Scores de benchmarks (LLM Stats)

(LLM Stats (zeroeval))

Agents

TAU3-Bench67.2%Aut.
MCP-Mark37.0%Aut.
VITA-Bench35.6%Aut.
Toolathlon26.9%Aut.
DeepPlanning25.9%Aut.

Code

ZClawBench52.6%Aut.
Claw-Eval50.0%Aut.
NL2Repo29.4%Aut.
SkillsBench28.7%Aut.

General

C-Eval90.0%Aut.

Language

MMLU-Redux93.3%Aut.
MMLU-Pro85.2%Aut.

Math

AIME 202692.7%Aut.
HMMT 202590.7%Aut.
HMMT2589.1%Aut.
MathVista-Mini86.4%Aut.
HMMT Feb 2683.6%Aut.
IMO-AnswerBench78.9%Aut.

Multimodal

VideoMMMU83.7%Aut.
VideoMME w/o sub.82.5%Aut.
MMMU81.7%Aut.

Reasoning

GPQANYU + Cohere + Anthropic (2023)86.0%Aut.
LiveCodeBench v680.4%Aut.
CharXiv-R78.0%Aut.
SWE-Bench Verified73.4%Aut.
SWE-bench Multilingual67.2%Aut.
SuperGPQA64.7%Aut.
MCP Atlas62.8%Aut.
Terminal-Bench 2.0Stanford × Laude Institute (2026)51.5%Aut.
SWE-Bench ProPrinceton NLP (2024)49.5%Aut.
Humanity's Last Exam21.4%Aut.

Search

WideSearch60.1%Aut.

Video

MLVU86.2%Aut.

Vision

MMBench-V1.192.8%Aut.
AI2D92.7%Aut.
RefCOCO-avg0.92 / 100Aut.
OmniDocBench 1.589.9%Aut.
VideoMME w sub.86.6%Aut.
RealWorldQA85.3%Aut.
EmbSpatialBench0.84 / 100Aut.
CC-OCR81.9%Aut.
MMMU-Pro75.3%Aut.
MVBench74.6%Aut.
LVBench71.4%Aut.
Hallusion Bench69.8%Aut.
RefSpatialBench0.64 / 100Aut.
SimpleVQA0.59 / 100Aut.
ODinW50.8%Aut.
ZEROBench-Sub0.34 / 100Aut.

Indices d'évaluation AA

(Artificial Analysis)

Aucune donnée d'évaluation AA disponible

Scores par catégorie LLM Stats

(LLM Stats (zeroeval))
Language
90
Structured Output
90
Biology
90
Long Context
80
Math
80
Multimodal
80
Physics
80
Spatial Reasoning
80
Embodied
80
Grounding
80
Healthcare
80
Chemistry
80
Text-to-image
80
Video
80
Legal
70
Reasoning
70
Finance
70
Frontend Development
70
General
70
Vision
70
Image To Text
60
Search
60
Economics
60
Code
50
Tool Calling
50
Agents
40

Tarification

Prix d'entrée$0.375 / 1M tokens
Prix de sortie$2.25 / 1M tokens
Prix mixte (3:1)$0.3125 / 1M tokens

Vitesse

Tokens/sec118.9
Délai du premier token2.02s
Temps de réponse47.39s

Classement des Prix par Fournisseur

Classement des Prix par Fournisseur

14 fournisseurs

Moins cher: engyPlus cher: Alibaba
FournisseurEntréeSortie
1engyMoins cher
$0.045
$0.3
2EmpirioLabs AI
$0.07
$0.42
3SaladCloud AI Gateway
$0.09
$0.6
4DeepInfra
$0.1
$0.95
5Venice AI
$0.1
$1
6ZeroGPU
$0.2
$1.5
7DevPass (LLM Gateway)
$0.248
$1.485
8Opper
$0.248
$1.485
9AIHubMix
$0.254
$1.524
10Cortecs
$0.279
$1.671
11302.AI
$0.28
$1.7
12Scaleway
$0.28038
$1.68225
13GreenPT
$0.342
$2.052
14AlibabaPRINCIPAL
$0.375
$2.25

Comparer les prix entre différents fournisseurs API pour ce modèle.

Sources externes