Qwen3.5

model
Alibaba
Fourth generation of the Qwen model family. Qwen3.5-397B-A17B native VLM built on Qwen3-Next architecture. 201 languages and dialects supported. 1M context window.
Blog PostGitHubArtificial Analysis

Outputs 5

Qwen3.5-397B-A17B

model

Qwen3.5-397B-A17B native VLM. 201 languages and dialects supported. 150k to 250k vocabulary expansion. 1M context window. Matches the >1T-parameter Qwen3-Max-Base.

Blog PostGitHubTechnical Report
Architecture MOE
Parameters 397B
Active params 17B
Context window 1,000,000
Languages 201

Variants

Name Parameters Notes
Qwen3.5-397B-A17B 397B —

Qwen3.5-Plus

model

Proprietary hosted iteration of the Qwen3.5-397B-A17B model. Outperforms Qwen3-Max on reasoning and coding while using less compute.

Blog Post
Architecture MOE

Qwen3.5 Medium Model Series

model

Medium-sized models in the Qwen3.5 family including agentic variants.

Announcement
Architecture MOE

Variants

Name Parameters Notes
Qwen3.5-35B-A3B 35B —
Qwen3.5-122B-A10B 122B Agentic
Qwen3.5-27B 27B Agentic, dense
Qwen3.5-Flash — Hosted production version optimized for latency

Qwen3.5-Max-Preview

model

Current state-of-the-art preview for the largest reasoning model.

Blog Post
Architecture MOE

Date approximate.

Qwen3.5 Small Model Series

model

Small dense models in the Qwen3.5 family.

Announcement
Architecture DENSE

Variants

Name Parameters Notes
Qwen3.5-0.8B 0.8B —
Qwen3.5-2B 2B —
Qwen3.5-4B 4B —
Qwen3.5-9B 9B —
multimodalmoeagenticnlpopen-weightreasoning