Qwen 3.7-Max released: Alibaba Agent model ranks among the top 6 in the world, and the open source ecosystem has made another breakthrough

Core conclusion

On May 20, 2026, the Alibaba Qwen team officially released Qwen 3.7-Max-Preview and Qwen 3.7-Plus-Preview. These two models are positioned as "Agent Frontier" - flagship models specifically optimized for AI Agent scenarios. On the LMSys Chatbot Arena list, Qwen 3.7-Max-Preview has ranked 6th in text capabilities and 5th in visual capabilities, competing head-on with the world's top models such as Google Gemini 3.5 Flash and Claude Opus 4.5.

Key Points

• Event Time: 2026-05-20
• Release content: Qwen 3.7-Max-Preview (Flagship level) and Qwen 3.7-Plus-Preview (Cost-performance version)
• Core positioning: Agent task optimization instead of traditional question and answer/text generation
• Arena Ranking: 6th Text, 5th Visual (LMSys Arena)
• Affected objects: AI Agent developers, open source model deployers, automated workflow builders

Background and trigger events

The release of Qwen 3.7-Max continues the high-frequency iteration rhythm of Alibaba Qwen team in 2026. The Qwen 3.5 series released earlier this year has received wide acclaim in the open source community for its MoE (Mixed Experts) architecture, and version 3.7 further points the finger at AI Agent, the hottest track in 2026.

Different from previous versions, Qwen 3.7-Max is not a pure question and answer model, but a reasoning model specially designed for Agent scenarios, emphasizing multi-step reasoning, tool invocation, code execution and autonomous decision-making capabilities.

Key Impact (by Dimension)

Adaptation suggestions

Based on the release of Qwen 3.7-Max, the following is a list of actions that can be implemented:

Short term (1-3 days)

• Test Qwen 3.7-Max-Preview on LMSys Arena and compare it with the currently used model
• Use Qwen 3.7 API to build a simple Agent prototype (tool call + multi-step reasoning)
• Compare the cost difference between Qwen 3.7-Plus and DeepSeek V4 Flash on the same tasks

Mid-term (1-2 weeks)

• Add Qwen 3.7 as a candidate model in the automated pipeline
• Incorporate Agent evaluation indicators (success rate, number of retries, token consumption) into model selection criteria
• Pay attention to the local deployment solution after the official version of Qwen 3.7-Max is open sourced

Long term (1 month+)

• Build a multi-model routing layer: Gemini 3.5 Flash for core decision-making, Qwen 3.7-Max for agent tasks, and Qwen 3.7-Plus for cost-sensitive tasks.
• Incorporate failed retries and observable logs of Agent tasks into pipeline standardization

Example: API call comparison

import openai # OpenAI

# Qwen 3.7-Max
client = openai.OpenAI(
    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
    api_key="YOUR_DASHSCOPE_API_KEY"
)

response = client.chat.completions.create(
    model="qwen3.7-max-preview",
    messages=[
 {"role": "system", "content": " AI Agent，。"},
 {"role": "user", "content": "，。"}
    ],
    tools=[{
        "type": "function",
        "function": {
            "name": "get_weather",
 "description": "",
            "parameters": {"type": "object", "properties": {"city": {"type": "string"}}}
        }
    }]
)

• HN ：Qwen 3.7-Max: The Agent Frontier
• Qwen （Twitter）

， tool_mentions Qwen、DeepSeek、Gemini、OpenAI、ChatGPT、Claude、n8n、Hermes Agent、LangGraph

Internal link guidance

• Want to know about open source large model deployment solutions? Watch: Qwen 3.7 Preview debuts in Arena: Alibaba Cloud’s large open source model evolves again, and local deployment welcomes new options
• Want to build your own AI Agent workflow? Watch: n8n AI Agent Node v2： Claude 4、Gemini DeepSeek R2
• Wondering how the Agent model differs from traditional models? Watch: Google I/O 2026 releases Gemini 3.5 Flash: The Agent era has officially arrived