Deploy and Use Alibaba’s New Qwen3‑Coder AI Programming Model on the Cloud

Model Introduction

On July 23, Alibaba Cloud AI platform released the open‑source AI programming large model Qwen3‑Coder. The model achieves state‑of‑the‑art results on Agentic Coding, Agentic Browser‑Use and foundational coding tasks. Its strongest variant, Qwen3‑Coder‑480B‑A35B‑Instruct, is a Mixture‑of‑Experts model with long‑context capability.

PAI‑Model Gallery Overview

PAI‑Model Gallery, a component of Alibaba Cloud PAI, aggregates high‑quality pretrained models from the open‑source community across LLM, AIGC, CV, NLP and other domains. It enables zero‑code end‑to‑end training, deployment and inference on the cloud.

Key features: zero‑code one‑click deployment, automatic cloud‑resource adaptation, ready‑to‑use APIs, full‑process operation management, and enterprise‑grade security with data staying in‑domain.

One‑Click Deployment of Qwen3‑Coder

Locate Qwen3‑Coder‑480B‑A35B‑Instruct in the Model Gallery or go directly to https://x.sm.cn/69wBHFA .

Click “Deploy”, select the required compute resources, and complete cloud deployment.

After deployment, retrieve the endpoint and token from the service page.

Use the model via API, PAI online debugging, or the WebUI.

Using Qwen3‑Coder in Agentic CLI

The open‑source Qwen Code tool (based on Gemini CLI) demonstrates how to call the deployed model. Install Qwen Code from https://github.com/QwenLM/qwen-code .

Configure the service by creating a .qwen/.env (or .env) file with the following environment variables:

export OPENAI_API_KEY="your_api_key_here"
export OPENAI_BASE_URL="your_api_base_url_here"
export OPENAI_MODEL="Qwen3-Coder-480B-A35B-Instruct"

Then test a case, e.g., generate a Snake game; the model produces correct logic and detailed planning.

Other Agentic Coding Tools

Qwen3‑Coder can also be used with Claude Code, Cline and other agentic programming tools by following similar configuration steps.

Additional Model Support

PAI‑Model Gallery continuously adds popular open‑source models, including full‑size Qwen3 models and the optimized DeepSeek‑R1, which offers up to 492% throughput improvement and significant latency reductions.

Contact

For model requests or support, join the PAI‑Model Gallery user group on DingTalk (79680024618).