Wow, this model was released on Reddit too and it's causing quite a stir.

Qwen-code was briefly released for free, and people who were using it well suddenly had the free tier end on April 15th, causing an uproar.

It seems public opinion has flipped overnight.

Original source: https://qwen.ai/blog?id=qwen3.6-35b-a3b

Following the launch of Qwen3.6-Plus, we are pleased to announce the open-source release of Qwen3.6-35B-A3B. This is a sparse yet remarkably performant Mixture-of-Experts (MoE) model with a total of 35 billion parameters but only 3 billion active parameters. Despite its superior efficiency, Qwen3.6-35B-A3B delivers outstanding agent coding performance that significantly outperforms its predecessor Qwen3.5-35B-A3B, and is comparable to much larger dense models such as Qwen3.5-27B and Gemma-31B. Supporting both multimodal thinking and non-thinking modes, Qwen3.6-35B-A3B has established itself as one of the most versatile open-source models currently available. Qwen3.6-35B-A3B is now available in real-time on Qwen Studio, can be called via API, and has been distributed with open weights for the community.

• Qwen3.6-35B-A3B is a fully open-source MoE model (total 35B / active 3B) with the following characteristics:

  • Exceptional agent coding capabilities competitive with much larger models

  • Strong multimodal perception and reasoning capabilities

• You can enjoy interactive chatting on Qwen Studio,
  call it as Qwen3.6-Flash on Alibaba Cloud Model Studio API (coming soon),
  or download and use the weights from Hugging Face and ModelScope.

Performance Evaluation

Below, we present comprehensive evaluation results comparing models of similar scale across diverse tasks and modalities.

Language Performance

With only 3 billion active parameters, Qwen3.6-35B-A3B surpasses Qwen3.5-27B, a dense model with 27 billion parameters, on several key coding benchmarks, and significantly outperforms its predecessor Qwen3.5-35B-A3B, especially in agent coding and reasoning tasks.

Note: Please refer to the footnotes at the bottom of the original text for evaluation settings and conditions for each benchmark.

Vision-Language Performance

Qwen3.6 is based on a native multimodal architecture, and Qwen3.6-35B-A3B demonstrates exceptional perception and multimodal reasoning capabilities that transcend scale with only approximately 3 billion active parameters. It exhibits performance comparable to Claude Sonnet 4.5 on most vision-language benchmarks, and in some tasks it has even recorded superior results. Particularly, it shows strength in the spatial intelligence field, achieving a score of 92.0 on RefCOCO and 50.8 on ODInW13.

Blank (--) indicates score not provided or not applicable.

Developing with Qwen3.6-35B-A3B

Qwen3.6-35B-A3B will soon be available on Alibaba Cloud Model Studio. Please wait a moment until complete preparation is finished.

Qwen3.6-35B-A3B is provided as open weights on Hugging Face and ModelScope for self-hosting, and can be called as qwen3.6-flash through the Alibaba Cloud Model Studio API. You can also try it immediately in Qwen Studio.

This model seamlessly integrates with major third-party coding assistants including OpenClaw, Claude Code, and Qwen Code, simplifying development workflows and enabling context-aware efficient coding experiences.

How to Use the API

This release supports the preserve_thinking feature: a feature that maintains thinking content from all previous turns in messages, and is recommended for agent tasks.

Alibaba Cloud Model Studio

Alibaba Cloud Model Studio supports industry-standard protocols and provides chat completions and responses APIs compatible with OpenAI specifications, as well as API interfaces compatible with Anthropic.

The example code for using the chat completions API is as follows:

"""
Environment variables (based on official documentation):
  DASHSCOPE_API_KEY: API key issued from https://modelstudio.console.alibabacloud.com
  DASHSCOPE_BASE_URL: (Optional) Base URL of the compatible mode API
    - Beijing: https://dashscope.aliyuncs.com/compatible-mode/v1
    - Singapore: https://dashscope-intl.aliyuncs.com/compatible-mode/v1
    - US (Virginia): https://dashscope-us.aliyuncs.com/compatible-mode/v1
  DASHSCOPE_MODEL: (Optional) Model name; can be overridden when using a different model
"""
from openai import OpenAI
import os

api_key = os.environ.get("DASHSCOPE_API_KEY")
if not api_key:
    raise ValueError(
        "DASHSCOPE_API_KEY is required. "
        "Set it via: export DASHSCOPE_API_KEY='your-api-key'"
    )

client = OpenAI(
    api_key=api_key,
    base_url=os.environ.get(
        "DASHSCOPE_BASE_URL",
        "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
    ),
)

messages = [{"role": "user", "content": "Introduce vibe coding."}]

model = os.environ.get("DASHSCOPE_MODEL", "qwen3.6-flash")

completion = client.chat.completions.create(
    model=model,
    messages=messages,
    extra_body={
        "enable_thinking": True,
        # "preserve_thinking": True,
    },
    stream=True
)

reasoning_content = ""  # entire reasoning trace
answer_content = ""     # entire response
is_answering = False    # whether answer phase has begun

print("\n" + "=" * 20 + "Reasoning" + "=" * 20 + "\n")

for chunk in completion:
    if not chunk.choices:
        print("\nUsage:")
        print(chunk.usage)
        continue

    delta = chunk.choices[0].delta
    
    # collect only reasoning content
    if hasattr(delta, "reasoning_content") and delta.reasoning_content is not None:
        if not is_answering:
            print(delta.reasoning_content, end="", flush=True)
        reasoning_content += delta.reasoning_content
    
    # receive content, begin answer phase
    if hasattr(delta, "content") and delta.content:
        if not is_answering:
            print("\n" + "=" * 20 + "Answer" + "=" * 20 + "\n")
            is_answering = True
        print(delta.content, end="", flush=True)
        answer_content += delta.content

For more information, please refer to the API documentation.

Coding and Agent Integration

Qwen3.6-35B-A3B has excellent agent coding capabilities and can seamlessly integrate with popular third-party coding assistants including OpenClaw, Claude Code, and Qwen Code.

OpenClaw

Qwen3.6-35B-A3B is compatible with OpenClaw (formerly Moltbot / Clawdbot), a self-hosted open-source AI coding agent. Connect to Model Studio and experience a complete agent coding environment in your terminal.

Getting Started Guide:

# Node.js 22+
curl -fsSL https://molt.bot/install.sh | bash   # macOS / Linux

# Set API key
export DASHSCOPE_API_KEY=

# Run OpenClaw
openclaw dashboard # web browser
# openclaw tui # start TUI in a new terminal

When using for the first time, edit the ~/.openclaw/openclaw.json file to configure OpenClaw to point to Model Studio. Find and create the following fields, then merge them — do not overwrite the entire file to preserve existing settings.

{
  "models": {
    "mode": "merge",
    "providers": {
      "modelstudio": {
        "baseUrl": "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
        "apiKey": "DASHSCOPE_API_KEY",
        "api": "openai-completions",
        "models": [
          {
            "id": "qwen3.6-flash",
            "name": "qwen3.6-flash",
            "reasoning": true,
            "input": ["text", "image"],
            "contextWindow": 131072,
            "maxTokens": 16384
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "modelstudio/qwen3.6-flash"
      },
      "models": {
        "modelstudio/qwen3.6-flash": {}
      }
    }
  }
}

Qwen Code

Qwen3.6-35B-A3B is designed for terminals and is compatible with Qwen Code, an open-source AI agent optimized for the Qwen series.

Getting Started:

# Node.js 20+
npm install -g @qwen-code/qwen-code@latest

# Start Qwen Code (interactive)
qwen

# Within the session:
/help
/auth

A login guide will be displayed on first use. You can switch authentication methods at any time using the /auth command.

Claude Code

The Qwen API also supports the Anthropic API protocol, so you can use it with tools like Claude Code to enjoy an enhanced coding experience:

# Install Claude Code
npm install -g @anthropic-ai/claude-code

# Set environment variables
export ANTHROPIC_MODEL="qwen3.6-flash"
export ANTHROPIC_SMALL_FAST_MODEL="qwen3.6-flash"
export ANTHROPIC_BASE_URL=https://dashscope-intl.aliyuncs.com/apps/anthropic
export ANTHROPIC_AUTH_TOKEN=

# Run CLI
claude

Summary

Qwen3.6-35B-A3B demonstrates that sparse MoE models can achieve outstanding agent coding and reasoning capabilities. With only 3 billion active parameters, it delivers performance comparable to dense models several times larger in active scale while demonstrating excellent results across multimodal benchmarks. As a fully open-source checkpoint, this model sets new standards achievable at its scale.

Moving forward, we will continue to expand the Qwen3.6 open-source family and tirelessly broaden the limits of what efficient and open models can achieve. We deeply appreciate community feedback and look forward to seeing what innovations you will create with Qwen3.6-35B-A3B. Additionally, the Qwen3.6 open-source family is continuing to expand, so we look forward to your attention to future releases!

Citation

If Qwen3.6-35B-A3B has been helpful, please cite the paper below:

@misc{qwen36_35b_a3b,
  title = {{Qwen3.6-35B-A3B}: Agentic Coding Power, Now Open to All},
  url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b},
  author = {{Qwen Team}},
  month = {April},
  year = {2026}
}

▶ Original source: https://qwen.ai/blog?id=qwen3.6-35b-a3b