Qwen3 VL 30B A3B🔒Private

• - 🧠 Mixture-of-experts vision-language model with about 3B active parameters.
• - 👁️ Unifies text generation with image and video understanding.
• - 📏 Serves a 128K-token context window for long multimodal inputs.
• - 🔒 Runs inside a Trusted Execution Environment with hardware attestation.
• - 🔧 Supports function calling and web search.
• - 🌐 Apache-2.0 licensed, released by Alibaba's Qwen team.
• - 🏢 Part of the broad Qwen3 multimodal lineup from Alibaba Cloud.

Alibaba

Alibaba Group is a Chinese multinational technology company founded in 1999 and headquartered in Hangzhou, Zhejiang. Originally built around e-commerce and cloud computing, Alibaba has become one of the most prolific contributors to open-weight AI research,…

Qwen3 VL 30B A3B is the multimodal branch of Alibaba's Qwen3 family, pairing a language backbone with a vision encoder so a single model can read text, images, and video. Its name reflects a mixture-of-experts design with roughly 3B parameters active out of 30B total, a layout the Qwen team uses across the Qwen3 series to keep inference efficient relative to capacity. This Venice deployment adds a privacy layer: the model runs inside a Trusted Execution Environment and exposes hardware attestation evidence for independent verification.

Compared with its same-family relatives, this checkpoint sits below the larger Qwen3 VL 235B, which applies the same vision-language approach at far greater scale. It also extends the text-only Qwen3 30B A3B: where that sibling handles language alone, the VL variant grafts on visual perception while retaining the comparable 30B-A3B architecture and a 128K context window.

Per the catalog configuration, the model supports vision input, function calling, and web search, making it suitable for document understanding, multi-image tasks, and tool-using agent workflows. As with any vision-language system, real-world accuracy on your specific images and documents is best confirmed through direct testing.