Qwen3.6-35B-A3B released!
This 35B parameter model matches coding performance of models 10x its active size and is fully open-source.
Alibaba's Qwen AI team has launched Qwen3.6-35B-A3B, a significant new entry in the open-source AI landscape. This model employs a sparse Mixture of Experts (MoE) architecture, which means it has a total of 35 billion parameters but only activates approximately 3 billion parameters for any given task during inference. This design makes it far more computationally efficient than dense models of similar capability. Crucially, it's released under the business-friendly Apache 2.0 license, allowing for commercial use without restrictive clauses.
The model's standout feature is its 'agentic coding' performance, which the team claims is on par with models that have 10 times its number of active parameters (around 30B). This makes it a highly cost-effective option for code generation and software development tasks. Beyond coding, Qwen3.6-35B-A3B is a multimodal model, meaning it can process and reason about both text and images. It also introduces a flexible 'multimodal thinking' mode, where users can choose between a more deliberate, chain-of-thought reasoning process or a faster 'non-thinking' mode for quicker responses, offering a trade-off between speed and accuracy.
The model is immediately available for developers to experiment with on several platforms, including the official Qwen Studio chat interface, Hugging Face, and ModelScope. This release continues Alibaba's strategy of pushing the frontier of open-source, commercially usable AI, providing a powerful and efficient alternative to closed models from companies like OpenAI and Anthropic for specific use cases like coding and multimodal analysis.
- Sparse MoE architecture with 35B total parameters but only 3B active, enabling high efficiency.
- Apache 2.0 licensed open-source model with agentic coding performance rivaling models with 30B active params.
- Multimodal capabilities with optional 'thinking' mode for complex reasoning and a faster 'non-thinking' mode.
Why It Matters
Provides a powerful, commercially-usable, and efficient open-source alternative for coding and multimodal AI tasks, challenging closed models.