Now in Foundry: Qwen3.5 Medium Model Series

2026-03-02 · Source: Microsoft Foundry Blog articles · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Intermediate, short

Summary

The Qwen3.5 Medium Model Series, now available in Microsoft Foundry, comprises three Vision Language Models (VLMs) featuring early-fusion multimodal training, a 262K native context window, and support for 201 languages under Apache 2.0. These models incorporate unified vision-language training, Gated Delta Networks for linear attention, and scalable reinforcement learning for post-training. The series includes a 27B dense model optimized for latency-sensitive applications, a 35B total parameter MoE model activating 3B parameters for high-throughput and cost-efficiency, and a 122B total parameter MoE model activating 10B parameters, offering frontier-class multimodal performance and expert-level knowledge depth. Each model is designed for specific use cases, from real-time visual inspection to complex financial research.

Key takeaway

For AI/ML Directors evaluating multimodal models for production, the Qwen3.5 Medium Series in Microsoft Foundry offers specialized options. Your choice should align with specific operational needs: Qwen3.5-27B for predictable low-latency tasks, Qwen3.5-35B-A3B for cost-optimized high-throughput scenarios, or Qwen3.5-122B-A10B for maximum capability in complex reasoning. Consider deploying directly via the Hugging Face collection in Foundry for streamlined integration and secure inference.

Key insights

Qwen3.5 models offer diverse VLM capabilities, from low-latency dense to cost-efficient MoE architectures.

Principles

• Early-fusion VLM training enhances reasoning.
• Gated Delta Networks improve inference throughput.
• MoE routing balances performance and cost.

Method

Qwen3.5 models use unified vision-language training, Gated Delta Networks for attention, and scalable RL across multi-agent environments for post-training.

In practice

• Use Qwen3.5-27B for real-time visual inspection.
• Deploy Qwen3.5-35B-A3B for high-throughput contract review.
• Leverage Qwen3.5-122B-A10B for complex financial analysis.

Topics

• Qwen3.5 Models
• Vision Language Models
• Mixture-of-Experts
• Multimodal AI
• Microsoft Foundry

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, Data Scientist

Related on AIssential

• Qwen3.5 Medium Models: Dense vs. MoE — Qwen3.5 medium models use Gated Deltanet for high throughput and low memory, showing robustness to quantization.
• What is trending in Hugging Face on Microsoft Foundry? Feb, 2, 2026 — Trending Hugging Face models are now production-ready in Microsoft Foundry, offering advanced AI capabilities.
• Introduction to Qwen3-VL — Qwen3-VL models offer advanced multimodal understanding and reasoning through architectural innovations and diverse task capabilities.
• 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.
• Now in Foundry: IBM Granite 4.1, NVIDIA Nemotron Nano Omni, and Qwen3.6-35B-A3B — New models in Microsoft Foundry offer diverse capabilities from multimodal reasoning to agentic coding and document intelligence.
• Can A Medium Qwen 3.5 Reason? Flash, 27B, 9B TEST — Smaller Qwen 3.5 models exhibit reduced causal reasoning and strategic planning capabilities on complex tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Microsoft Foundry Blog articles.