Qwen3.5-397B-A17B: the smallest Open-Opus class, very efficient model

2026-02-16 · Source: AINews · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Advanced, extended

Summary

Alibaba has released Qwen3.5-397B-A17B, an open-weight model featuring native multimodality, spatial intelligence, and a hybrid linear attention + sparse MoE architecture supporting 201 languages and long context windows up to 256K tokens. The model demonstrates improvements over previous versions like Qwen3-Max and Qwen3-VL, with a sparsity ratio of about 4.3%. Community discussions highlight the Gated Delta Networks enabling efficient inference despite the large model size (~800GB BF16), with successful local runs reported on Apple Silicon using quantization techniques. The hosted API version, Qwen3.5-Plus, supports 1M context and integrates search and code interpreter features. This release follows other Chinese labs like Z.ai, Minimax, and Kimi in refreshing large models. The model is licensed under Apache-2.0 and is anticipated to be the last major release before DeepSeek v4. Additionally, Pete Steinberger has joined OpenAI.

Key takeaway

For AI architects evaluating large language models for deployment, Qwen3.5-397B-A17B presents a compelling open-weight option with native multimodality and efficient sparse MoE architecture. Its reported ability to run locally on systems like Apple Silicon with quantization, despite its ~800GB BF16 scale, suggests a practical path for integrating advanced capabilities without exclusive reliance on cloud APIs. You should investigate its performance on your specific multimodal and long-context tasks, especially considering its Apache-2.0 license.

Key insights

Qwen3.5 offers multimodal, long-context capabilities with efficient sparse MoE architecture, enabling local deployment despite its large scale.

Principles

• Sparse MoE architectures can enable efficient inference for large models.
• Quantization techniques facilitate local deployment of frontier models.

Method

Qwen3.5 utilizes Gated Delta Networks and a hybrid linear attention + sparse MoE architecture to manage long contexts and achieve efficient inference.

In practice

• Consider Qwen3.5 for multimodal applications requiring long context windows.
• Explore 4-bit quantization for running large models like Qwen3.5 on consumer hardware.

Topics

• Qwen 3.5
• Large Language Models
• AI Agents
• Multimodal AI
• AI Infrastructure

Code references

• Healthy-Nebula-3603/gpt5.2-codex_xhigh-proof-of-concept-GBA-emulator-in-assembly-
• EleutherAI/gpt-neox
• MrMeatikins/planbot-resource
• sandover/ergo
• Archelunch/dspy-repl

Best for: Machine Learning Engineer, AI Architect, NLP Engineer, AI Engineer, AI Researcher, MLOps Engineer

Related on AIssential

• Alibaba Qwen Team Releases Qwen3.5-397B MoE Model with 17B Active Parameters and 1M Token Context for AI agents — Qwen3.5 MoE model achieves 400B-class reasoning with 17B-level inference speed and a 1M token context.
• MiniMax M3 IS INSANE! BEST Opensource AI Model! Beats Opus 4.7 and 50x Cheaper! (Fully Tested) — MiniMax M3 is an open-weight, natively multimodal model offering frontier performance in coding and agentic tasks at significantly lower costs.
• Qwen 3.5 The GREATEST Opensource AI Model That Beats Opus 4.5 and Gemini 3? (Fully Tested) — Qwen 3.5 is a powerful open-source, multimodal AI model excelling in speed and diverse applications.
• Qwen3-Coder-Next: How to Run Qwen’s 80B Sparse MoE Coder — Qwen3-Coder-Next uses sparse MoE and hybrid attention to enable efficient, long-context coding agent capabilities.
• 🗞️ Alibaba released their open-source Qwen3.5-9B model, and it holds its own against OpenAI's 120B param gpt-oss system. — Smaller, efficient AI models are achieving performance comparable to much larger systems through architectural innovations and multimodal training.
• The Sequence AI of the Week #818: You Cannot Miss Qwen 3.5 — The Alibaba Qwen team recently launched the Qwen 3.5 series, featuring a flagship Qwen3.5-397B-A17B model that competes with proprietary giants like GPT-5.2 and Claude Opus 4.5 on benchmarks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AINews.