NEW Qwen 3 Coder: FULLY FREE AI Coder! Develop a Full-stack App Without Writing ANY Code! (FREE API)

Quen 3 Model: Open-Source Coding Powerhouse

Key Concepts:

• Quen 3: Alibaba's new open-source Mixture of Experts (MoE) model.
• Active Parameters: The subset of parameters within an MoE model that are used for a specific task.
• Context Length: The amount of text a model can consider when generating a response.
• Client: An autonomous coding agent integrated within an IDE.
• Open Router: A platform for accessing various AI models through a unified API.
• Ollama: A tool for running open-source models locally.
• p5.js: A JavaScript library for creative coding, focused on visualization and interaction.
• SVG: Scalable Vector Graphics, an XML-based vector image format.

1. Introduction to Quen 3

Alibaba has released the Quen 3 model, featuring two open-source Mixture of Experts (MoE) models:

• Quen 3 235B: 235 billion parameters, with 22 billion active parameters.
• Quen 3 30B: A lightweight version with 30 billion parameters and 3 billion active parameters.

In addition to the MoE models, six dense models ranging from 0.6 billion to 32 billion parameters were released. These models are available under the Apache 2.0 license via Hugging Face and are optimized for 128K or 32K context lengths.

2. Performance Benchmarks

The Quen 3 235B model demonstrates impressive performance, matching or exceeding top-tier models like DeepSeek R1 in various categories, including math, coding, and reasoning. It outperforms DeepSeek R1, Grok-1.5 Beta, Gemini 1.5 Pro, and OpenAI's GPT-3.5 Mini and 01 models.

3. Integrating Quen 3 with an AI Coding Agent (Client)

The video focuses on leveraging Quen 3's coding capabilities by integrating it with an autonomous coding agent called Client. Client is an IDE extension that can create, edit files, execute commands, and browse the web autonomously (with user permission).

4. Setting up Client with Quen 3

Step-by-step process:

1. Install Client Extension: Install the Client extension in your IDE (e.g., VS Code, Vinsurf, Cursor).
2. Open Router API Key (Optional):
   • Create a free account on Open Router.
   • Set up an API key.
   • In Client's settings, select "Open Router" as the AI provider.
   • Paste the API key and select the Quen 3 model.
3. Local Hosting with Olama (Alternative):
   • Install Olama.
   • Search for and install the desired Quen 3 model size (excluding the 235B model due to resource constraints).
   • Run the model in the terminal using the command provided by Olama.
   • In Client's settings, select "Ollama" as the provider.
   • Enter the model ID, base URL, and model context window.

5. Testing Quen 3's Coding Capabilities

Several prompts were used to evaluate Quen 3's coding abilities:

• SAS Landing Page: Quen 3 generated a modern-looking SAS landing page. The generation cost approximately 1 cent using Open Router's pricing. The presenter considered the result better than DeepSeek R1's output for the same task.
• TV with 0-9 Channels: Quen 3 created a TV simulation with animated channel transitions, demonstrating creativity and visual engagement. The model used spatial masking logic to create different channels. The presenter considered this one of the best generations for this prompt.
• Butterfly Representation in SVG: Quen 3 initially produced an imperfect butterfly SVG. After a second instruction to fix the first generation, it generated a good-looking butterfly. This was a better result than previous attempts with other models.
• Animated Weather Card: Quen 3 created a basic animated weather card displaying temperature, weather type (sunny, cloudy, rainy, snowy), and a dynamic animated icon. The UI was considered basic, but the model captured the core requirements.

6. Conclusion

The Quen 3 model is presented as a compelling open-source coding model, particularly suitable for local use. Its coding performance is highlighted, making it a valuable tool for developers. The video encourages viewers to try the model and provides links to resources for getting started.