Qwen 3.6 35B-A3B

This is a self-hosted deployment of the Qwen 3.6 35B-A3B  large language model. It runs as a single GPU-powered EC2 instance allowing you to keep your data private and leverage unlimited tokens. Access to the model is via HTTPS, ensuring data is encrypted in-transit at all times. Highlights of the Qwen 3.6 35B model include:

• Utilizes a sparse Mixture-of-Experts (MoE) architecture with 35 billion total parameters and only 3 billion active parameters to deliver flagship-level performance with massive efficiency.

• Features exceptional agentic coding capabilities that surpass much larger dense models and excel at repository-level reasoning and terminal-based workflows.

• Introduces the "preserve thinking" feature, which retains reasoning traces from all preceding conversation turns to improve consistency in complex, multi-step tasks.

• Native multimodal support enables high-performance perception and reasoning across text, images, and video, particularly in tasks requiring spatial intelligence.

• Supports a standard context window of 262K tokens, which can be extended up to 1 million tokens using specialized techniques like YaRN.

• Incorporates Multi-Token Prediction (MTP) to enable speculative decoding, significantly increasing inference speed for structured data and code generation.

• Provides flexible interaction through a native "Thinking Mode" for deep reasoning and a "Non-Thinking Mode" for instant, direct responses.

• Fully open-source under the Apache 2.0 license, allowing for unrestricted commercial use and local deployment on consumer-grade hardware.