Qwen3.6-35B-A3B - Alibaba Tongyi's open-source Hybrid Expert MoE Model

What is Qwen3.6-35B-A3B?

Qwen3.6-35B-A3B is an open-source hybrid expert (MoE) model launched by Alibaba's Tongyi Qianwen team, with a total of 35 billion parameters and 3 billion activation parameters. The model emphasizes extreme inference efficiency and agent programming capabilities, outperforming the denser model Qwen3.5-27B with a larger parameter scale in multiple programming benchmark tests, and significantly outperforming Google's latest Gemma 4 series models. As the first open-source version of the Qwen3.6 series, it natively supports multimodal perception and inference, and its visual language capabilities are on par with or even surpass those of Claude Sonnet 4.5, making it one of the most versatile open-source models currently available.

Main functions of Qwen3.6-35B-A3B

• Intelligent agent programmingThe model possesses excellent agentic coding capabilities, performs exceptionally well on programming benchmarks such as SWE-bench and Terminal-Bench, and can be seamlessly integrated into third-party programming assistants such as OpenClaw, Claude Code, and Qwen Code.
• Multimodal perception and reasoningIt natively supports visual language understanding and performs outstandingly on visual question answering benchmarks such as MMMU and RealWorldQA, with significant advantages in spatial intelligence (RefCOCO 92.0, ODINW13 50.8).
• Dual-mode reasoningIt supports flexible switching between thinking mode (complex reasoning) and non-thinking mode (rapid response) to adapt to different task scenarios.
• Efficient ReasoningIt adopts a sparse MoE architecture, which can achieve performance comparable to dense models of several times the size by activating only 3 billion parameters, thus greatly reducing inference costs.
• Tool Invocation and MCP SupportIt supports function calls, code interpreters, and MCP (Model Context Protocol), and can connect to external tools and APIs to complete complex tasks.
• Long context processingIt supports 200K context windows and can handle tasks such as long document comprehension and long-process code generation.

How to use Qwen3.6-35B-A3B

• Online experience:Visit the Qwen Studio website (https://chat.qwen.ai/) to start a conversation without registration. It supports text and image input and allows you to instantly experience the programming and multimodal capabilities of your models.
• API calls (production deployment)
  • Alibaba Cloud Hundred Refinement PlatformLog in to the Alibaba Cloud Refinement Console and select the model. qwen3.6-flash Create an API Key. Supports standard OpenAI protocols (Chat Completions/Responses API) and Anthropic protocols, seamlessly replacing existing GPT/Claude interfaces. It is recommended to enable this feature when making calls. preserve_thinking Parameters are set to preserve thought processes and optimize agent task performance.
  • Local deploymentDownload open-source weights from Hugging Face and load them using vLLM, Ollam, or Transformers.
• Third-party tool integration
  • OpenClawIn the configuration file, set the Base URL to the Alibaba Cloud Bailian endpoint, enter the API Key, and select the model name. qwen3.6-flashThis allows you to use the model in the terminal to replace the default encoding assistant.
  • Qwen CodeAn open-source terminal AI agent optimized specifically for the Qwen series, with initial input... /auth After completing the Alibaba Cloud Hundred Refinements Certification, the system automatically identifies and calls Qwen3.6-35B-A3B for code generation and tool invocation.
  • Claude CodeBecause the API is compatible with the Anthropic protocol, the endpoint address and model name can be directly replaced in the Claude Code configuration to obtain a coding experience with visual capabilities.

Key information and usage requirements for Qwen3.6-35B-A3B

• Open source licenseThe model weights are completely open source, supporting local deployment and commercial use, and can be downloaded from the Hugging Face and ModelScope platforms.
• Online experienceQwen Studio allows for direct interactive dialogue, providing full functionality without the need for deployment.
• API callsSoon, support will be provided for Alibaba Cloud's Hundred Refinements API, with the call name being...qwen3.6-flashIt is compatible with the OpenAI specification's Chat Completions and Responses API, as well as the Anthropic API protocol.
• Hardware RequirementsCompared to dense models, it significantly lowers the barrier to local deployment, requiring only enough video memory to support 3 billion activation parameters to run, making it suitable for local deployment by individual developers.
• Third-party integrationIt is already compatible with mainstream AI programming assistants such as OpenClaw (Moltbot), Qwen Code, and Claude Code, and can be directly integrated into existing development workflows.
• Special featuresAPI supportpreserve_thinkingParameters can retain the thought content of previous rounds in the message, and are recommended for complex agent tasks.

The core advantages of Qwen3.6-35B-A3B

• Extreme parameter efficiencyThe 3B activation parameter can surpass the 27B dense model in programming and inference tasks, achieving inference economy with "small size and big power".
• Top-notch intelligent agent programming capabilitiesIt outperforms models of similar or even larger scale on authoritative programming benchmarks such as SWE-bench Verified (73.4) and Terminal-Bench 2.0 (51.5).
• Visual ability benchmarked against closed-source modelsIts multimodal performance is on par with Claude Sonnet 4.5, with particularly outstanding spatial intelligence (RefCOCO 92.0) and document understanding capabilities.
• Full-scenario ecosystem compatibilityIt supports both OpenAI and Anthropic API protocols, allowing for seamless replacement of models in existing workflows and reducing migration costs.

Project address of Qwen3.6-35B-A3B

• HuggingFace model libraryhttps://huggingface.co/Qwen/Qwen3.6-35B-A3B

Comparison of Qwen3.6-35B-A3B with similar competing products

Application scenarios of Qwen3.6-35B-A3B

• AI-assisted programmingIt automates code generation, bug fixing, code refactoring, and code review, and can be integrated into IDEs as an intelligent programming assistant.
• Intelligent agent development: Build autonomous agents that can call tools, browse web pages, and execute code for automated operation and maintenance, data analysis and other workflows.
• Multimodal content understandingIt is designed to handle tasks such as document analysis with mixed text and images, chart interpretation, visual question answering, and video content understanding.
• End-side and edge deploymentDue to its low activation parameters and high inference efficiency, it is suitable for deployment on resource-constrained edge devices or edge servers to provide localized AI services.
• Educational ResearchIt serves as an open-source foundation model for academic research, algorithm teaching, or domain-specific fine-tuning training.