Qwen3.5-397B-A17B — reviews, specs & pricing

Alibaba's first Qwen3.5 release: a 397B-parameter MoE with 17B active parameters, Apache 2.0 licensed, hosted as Qwen3.5-Plus on Model Studio.

Summary

Alibaba's Qwen3.5-397B-A17B is the debut model of the Qwen3.5 family, leveraging a massive Mixture of Experts architecture with 397 billion total and 17 billion active parameters. Released under the permissive Apache 2.0 license, it excels in both conversational and programming tasks. Users can run it locally or access it as Qwen3.5-Plus on Alibaba's Model Studio cloud platform.

Sample use case

A software development agency can deploy Qwen3.5-397B-A17B on-premises to build an automated code review and pair-programming assistant for their development team. By leveraging the model's chat and coding capabilities, developers can safely input proprietary codebases to identify bugs, generate boilerplate code, and receive real-time architecture recommendations. The model's active-parameter efficiency ensures low-latency responses, making the internal tool highly responsive without requiring prohibitive computing budgets.

Specifications

• Provider: alibaba
• License: open
• Parameters: 397B
• Released: 2026-02-17

Pros

• Permissive Apache 2.0 license for commercial use
• Highly efficient MoE design with only 17B active parameters
• Optimized for advanced chat and code generation tasks
• Conveniently hosted on Model Studio as Qwen3.5-Plus

Cons

• Massive 397B footprint requires significant disk storage for local deployment
• High total parameter count demands substantial RAM/VRAM even with MoE active limits
• First release of the 3.5 generation may have minor optimization gaps

Average rating 0.0 from 0 community reviews on Reviuws.