Qwen3.7 Plus: Multimodal 1M-Context Model Guide

Last verified: August 4, 2026. This is an independent technical guide and is not operated by or affiliated with Alibaba Cloud or the Qwen team. Confirm availability, billing, and regional requirements in the provider documentation before deployment.

Qwen3.7 Plus is Alibaba Model Studio’s recommended balanced Qwen model for general-purpose applications. It combines a one-million-token context window with text, image, and video understanding, hybrid thinking, Function Calling, built-in tools, and text output.

Its API model ID is qwen3.7-plus. It is the current balanced starting point for broad multimodal and agent workflows. Use qwen3.7-flash when cost and throughput dominate, and evaluate the flagship qwen3.8-max when the strongest capability matters more than price.

Qwen3.7 Plus specifications

Specification	Verified information
Rolling model ID	qwen3.7-plus
Fixed snapshot	qwen3.7-plus-2026-05-26
US-scope model ID	qwen3.7-plus-us
Input	Text, images, and video
Output	Text
Context window	Up to 1,000,000 tokens
Maximum output	Official-source discrepancy checked August 4, 2026: the QwenCloud Marketplace card displays 131K, while the text and vision developer matrices display 64K. Treat this as route-specific and configure conservatively until tested.
Maximum images	Up to 2,048 for documented URL or path workflows; Base64 requests are capped at 250
Maximum videos	Up to 64
Video limit	Up to 2 hours and 2 GB
Maximum image size	Up to 16 million pixels per image
Thinking	Hybrid; enabled by default
Function Calling	Supported
Structured output	Supported, with non-thinking mode preferred for reliable JSON

Why choose Qwen3.7 Plus?

Qwen3.7 Plus is the sensible starting point when an application needs broad capability at the balanced tier. Alibaba recommends it for chatbots, content generation, summarization, document processing, coding tools, and office-productivity workflows. Compare it with qwen3.7-flash for lower-cost work and qwen3.8-max for a new highest-capability evaluation.

• Long-document and large-codebase analysis using the 1M context window.
• Image understanding, visual question answering, and information extraction.
• Video understanding within the documented duration and file-size limits.
• Agents that combine reasoning with custom functions or supported built-in tools.
• Multistep analysis using thinking mode.
• Lower-cost direct responses using non-thinking mode.

These are appropriate workload categories, not performance guarantees. Test the model with representative prompts, document formats, languages, tool definitions, and failure cases before selecting it for production.

Model alias and fixed snapshot

On the verification date, Alibaba’s pricing documentation mapped qwen3.7-plus to qwen3.7-plus-2026-05-26. The rolling alias is convenient for applications that accept provider-managed updates. The dated snapshot provides a more stable target for regression testing and controlled releases.

The separate qwen3.7-plus-us identifier is associated with the US deployment scope. A model ID, API key, workspace, and base URL must be selected from compatible deployment scopes; these values should not be mixed across regions.

Text, image, and video understanding

Qwen3.7 Plus accepts text, images, and video and returns text. It can therefore describe or analyze visual content, extract information, answer questions about images, and reason over supported video inputs. It is not documented as a native image-generation, video-generation, or audio-output model.

The published limits include up to 2,048 image items in documented URL or local-path workflows, up to 250 Base64 images, and up to 64 videos. Images can be up to 16 million pixels each. Video inputs can be up to two hours or 2 GB, subject to the provider’s file and request rules.

The QwenCloud developer matrices checked August 4, 2026 pair the 1M context category with a 64K maximum-output label, while the model-specific Marketplace card displays 131K. Under either documentation path, input, reasoning, tools, and output share the applicable request budget; do not advertise one million input tokens plus an unrestricted extra response.

Modality support can differ by API. Check the selected Chat Completions, Responses, Anthropic-compatible, or DashScope interface before implementing image or video input.

Hybrid thinking

Qwen3.7 Plus supports thinking and non-thinking operation in the same model family. Thinking is enabled by default.

• Use thinking for complex reasoning, architecture planning, debugging, and multi-document comparisons.
• Disable thinking for direct extraction, formatting, classification, and other tasks where extended reasoning is unnecessary.
• Use enable_thinking with Chat Completions.
• Use reasoning.effort with the Responses API.
• Use streaming for a better user experience with long reasoning requests.

The API separates reasoning content from the final answer. Reasoning tokens are billed as output. Retaining previous reasoning with preserve_thinking also adds that retained content to later input billing.

Function Calling and built-in tools

Qwen3.7 Plus supports Function Calling, including workflows where a model selects from functions defined by the application. The application remains responsible for authorizing and executing each function.

Tool arguments must be validated before they are used with databases, files, private APIs, payments, or other sensitive systems.

Alibaba also documents built-in tools such as web search, code execution, and web extraction for supported configurations. Tool availability and additional charges can depend on the API, deployment scope, and account plan. Do not state that every request has web access or that all built-in tools are included without charge.

Structured JSON output

Structured output can be enabled with response_format: {"type":"json_object"}. The prompt must explicitly contain the word JSON and describe the required fields.

JSON mode is not the same as guaranteed strict JSON Schema enforcement. For reliable automation, disable thinking, parse the returned JSON, validate it against an application-owned schema, and handle failures with a controlled retry or repair step.

Qwen3.7 Plus pricing

The table below uses QwenCloud’s standard list prices verified on August 4, 2026. The pricing tier is selected by the total input tokens in one request, and all tokens in that request use the selected tier.

Input tokens per request	List input price	List output price
More than 0 and up to 256K	US$0.40 per 1M tokens	US$1.60 per 1M tokens
More than 256K and up to 1M	US$1.20 per 1M tokens	US$4.80 per 1M tokens

Temporary promotion observed: the Qwen3.7 Plus Marketplace card displayed 20% off on August 4, including US$0.32 input and US$1.28 output beside the first-band US$0.40 / US$1.60 list prices. This is a dated offer, not the permanent comparison baseline; confirm the current card and account at purchase time.

An eligible deployment may receive a limited free quota, and Batch can have separate discounts and context limits. Do not combine promotions, quota, cache, or Batch assumptions unless the exact billing documentation permits it.

Node.js multimodal API example

This OpenAI-compatible Chat Completions example sends text and one image. Replace the example image URL with a file your application is authorized to process. Configure DASHSCOPE_BASE_URL for the same region and workspace as the API key.

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.DASHSCOPE_API_KEY,
  baseURL: process.env.DASHSCOPE_BASE_URL,
});

const stream = await client.chat.completions.create({
  model: "qwen3.7-plus",
  messages: [
    {
      role: "system",
      content:
        "Analyze the supplied image carefully. Separate observations from conclusions.",
    },
    {
      role: "user",
      content: [
        {
          type: "text",
          text:
            "Explain this system diagram and identify three operational risks.",
        },
        {
          type: "image_url",
          image_url: {
            url: "https://assets.example.com/system-diagram.png",
          },
        },
      ],
    },
  ],
  enable_thinking: true,
  stream: true,
  stream_options: {
    include_usage: true,
  },
});

for await (const chunk of stream) {
  const answer = chunk.choices[0]?.delta?.content;
  if (answer) process.stdout.write(answer);
}

For reproducible production tests, change the model ID to qwen3.7-plus-2026-05-26. Add input validation, file-type checks, size limits, timeouts, usage logging, and output moderation according to the application’s risk level.

Limitations and unsupported claims

• Qwen3.7 Plus is not documented as an open-weight or self-hosted model.
• No official parameter count, expert count, training-token total, or local GPU requirement is published for this model ID.
• Its documented output is text; do not claim native image, video, or audio generation.
• Audio input and output are not listed among its supported modalities.
• Built-in tool support does not mean every tool is available through every API and region.
• Function Calling does not execute custom tools without application code.
• JSON mode should not be presented as guaranteed strict JSON Schema compliance.
• Free quotas, rate limits, and deployment scopes are not universal.
• Do not publish temporary promotional pricing as the permanent model price.

Qwen3.7 Plus FAQ

Does Qwen3.7 Plus support one million tokens?

Yes. Alibaba publishes a context window of up to one million tokens. Input, conversation history, reasoning, tool results, and generated output must remain within the applicable request budget.

Can Qwen3.7 Plus process images and video?

Yes. Its documented inputs are text, images, and video, while its output is text. File, item-count, duration, and interface restrictions still apply.

Does Qwen3.7 Plus generate images?

Not as a native model output. Its documented output modality is text. A separately provided image-generation tool is a platform tool, not native image output from Qwen3.7 Plus.

Is Qwen3.7 Plus better than Qwen3.7 Max?

They target different priorities within the Qwen3.7 series: Qwen3.7 Max is the higher reasoning tier, while Plus balances capability and cost and has clearly documented image and video input support. Across the current hosted lineup, qwen3.8-max is the overall flagship.

Is Qwen3.7 Plus free?

It is a paid token-based Model Studio service. Eligible International activations may receive a limited one-million-token quota valid for 90 days, subject to Alibaba’s quota conditions.

Official references

• Qwen3.7 Plus Marketplace card: route limits and current offer
• QwenCloud text-model matrix
• QwenCloud vision-model matrix
• QwenCloud standard PAYG list prices
• Qwen3.7 Plus announcement

Sources were checked August 4, 2026. This page does not claim an independently tested maximum output, promotion duration, latency, or benchmark result. Explore the Qwen model directory, API guide, or pricing reference.