Qwen 3.7 Plus
Alibaba's cost-effective multimodal agent model — strong in agentic workflows, vision-language tasks, and GUI automation at low cost.
Overview
What is Qwen 3.7 Plus
Qwen 3.7 Plus is Alibaba Cloud's multimodal agent model, released June 2, 2026. It extends the Qwen 3.7 text backbone with vision and video understanding, enabling GUI interaction, code generation from visual references, and end-to-end mobile app navigation — all while returning text output.
Running it privately on Venice
On Venice, Qwen 3.7 Plus runs with anonymized privacy — your prompts are not stored or profiled. This enables secure, uncensored use of its full agentic capabilities, including vision, tool use, and web search, without surveillance. You maintain sovereignty over inputs while leveraging a powerful, production-ready agent model.
Assessment
Strengths and limitations
- Highly cost-effective: priced at roughly one-sixth of Qwen3.7-Max and significantly lower than many frontier models.
- Strong agentic performance with a 67th percentile ranking in tool use and autonomous workflows (τ²-bench: 93.0).
- Advanced GUI interaction: reads screens, grounds clicks, and navigates apps end-to-end with ScreenSpot Pro 79.0.
- Supports multimodal inputs: text, images, and video — for real-world scene understanding and visual QA.
- Integrated web search, function calling, and structured output for complex, real-time workflows.
- Proprietary and API-only: no open weights, so self-hosting or fine-tuning is not possible.
- Lower raw reasoning and knowledge benchmarks compared to Gemini 3.1 Pro across multiple evaluations.
- Max output limited to 32,768 tokens despite 1M context window, constraining long-form generation.
Capabilities
What it supports
- Tool use / function calling
- Vision (image input)
- Reasoning
- Web search
- Code-optimized
- Structured output (JSON schema)
- Audio input
- Video input
- Multiple image inputs
- Log probabilities
Specifications
Datasheet
- Maker
- Alibaba Cloud
- Released
- June 2, 2026
- Architecture
- Multimodal vision-language agent based on Qwen 3.7
- Parameters
- Not disclosed
- Open weights
- No — proprietary
- Context window
- 1,000K tokens
- Max output
- 65.536K tokens
- Capabilities
- Vision, Function calling, Reasoning, Web search, Code-optimized
- Privacy on Venice
- Anonymized — prompts not stored
- Available on Venice since
- Jun 2026
- License
- Proprietary
API
Call it from your code
Venice exposes an OpenAI-compatible API. Point your base URL at Venice and pass the model id.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen-3-7-plus",
"messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
}'Pricing
What it costs on Venice
Billed per token on Venice: $0.50 per 1M input tokens and $2 per 1M output tokens.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Alternatives
How it compares
| Model | Max resolution | Strongest at | Open weights | Price (Venice) |
|---|---|---|---|---|
| Qwen 3.7 Plus | N/A (text output only) | Agentic workflows, GUI automation | No | $0.50 in · $2 out / 1M |
| Claude Opus 5 | N/A | Reasoning, accuracy | No | $6 in · $30 out / 1M |
| DeepSeek V4 Flash 0731 | N/A | Speed, low cost | No | $0.17 in · $0.35 out / 1M |
| Google Gemma 4 31B Instruct | N/A | Open weights, efficiency | Yes | $0.12 in · $0.36 out / 1M |
Top-tier GUI grounding and cost efficiency for agent tasks.
Use cases
What it is good for
- 01Automating GUI-based workflows like browser navigation, app testing, and desktop software control.
- 02Generating code from screenshots or visual designs using grounded understanding.
- 03Multimodal customer support agents that interpret user-submitted images or videos.
- 04Real-time research workflows combining web search, structured output, and tool use.
- 05Low-cost agentic automation for startups and developers needing vision and reasoning.
Prompting
Getting better results
Include clear visual references when asking for code or UI descriptions — the model excels at grounding.
Use function calling to chain actions: e.g., 'search this topic, summarize results, then draft an email'.
Enable web search for up-to-date answers — it integrates seamlessly with reasoning steps.
Structure complex tasks step-by-step — the agent performs best with clear, modular instructions.
Version history
Text-only foundation model
Current — adds vision, video, agent capabilities
FAQ
Frequently asked questions
Qwen 3.7 Plus is a multimodal agent model by Alibaba Cloud, released on June 2, 2026. It combines strong text reasoning with vision and video understanding to enable GUI interaction, code generation from visual inputs, and autonomous task execution — all returning text output.
On Venice, Qwen 3.7 Plus is billed at $0.50 per 1M input tokens and $2 per 1M output tokens. Cached input is even cheaper at $0.05 per 1M tokens, making repeated context reuse highly efficient.
No. Qwen 3.7 Plus is a proprietary, API-only model from Alibaba Cloud. It is not open source, and no open weights have been released. You can use it via paid API access on platforms like Venice.
Yes. Qwen 3.7 Plus supports image and video input, enabling it to read screens, interpret visual scenes, and ground actions in GUIs. It is a perception model — it understands visuals but returns text, not generated images.
Yes. Qwen 3.7 Plus supports function calling and tool use, allowing it to integrate with external systems, automate workflows, and perform multi-step agent tasks like web search, code execution, and API interactions.
Qwen 3.7 Plus has a 1-million-token context window, allowing it to process extremely long documents, codebases, or multimodal inputs in a single session.
Gemini 3.1 Pro leads in most reasoning and knowledge benchmarks, but Qwen 3.7 Plus is about 10x cheaper per token and stronger in GUI automation tasks like ScreenSpot Pro. Choose Gemini for raw capability, Qwen for cost-efficient agentic workflows with vision.
No. Qwen 3.7 Plus is a perception model — it can read and understand images and video, but it only returns text output. It does not generate images.
While Qwen 3.7 Plus is not labeled as uncensored by its maker, running it on Venice ensures your prompts are not stored or profiled. This provides a privacy layer, though the model itself may still apply content filters per Alibaba's policies.
Run Qwen 3.7 Plus privately
No prompt logging. No data used for training.