Alibaba Releases Qwen 3.5 Multimodal Models
Alibaba's Qwen released two Qwen 3.5 models on Feb 15, 2026: an open-weight MoE Qwen3.5-397B-A17B and a proprietary Qwen3.5 Plus API. The open model uses a hybrid linear-attention and sparse mixture-of-experts design that activates 17 billion of 397 billion parameters for inference efficiency, with weights posted to Hugging Face (807GB) and GGUF variants available. The Plus API supports 1M-token context, search, and a code interpreter.
Key Points
- 1Launches two Qwen 3.5 models: open-weight MoE (Qwen3.5-397B-A17B) and proprietary Plus API.
- 2Implements hybrid linear-attention and sparse MoE to activate 17B of 397B parameters for efficiency.
- 3Provides practitioners open weights (807GB) and hosted Plus with 1M-token context and tools.
Scoring Rationale
Strong product relevance and accessibility, boosted by open weights and 1M context, limited by sparse benchmarking and brief coverage.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
