Alibaba Launches Qwen3.7-Flash, a Lightweight, Low-Cost Vision-Language Model with Stronger Agent Skills

Alibaba has released “Qwen3.7-Flash,” positioned as the lightweight, low-cost tier of the Qwen3.7 series. The series already includes Qwen3.7-Max (May 2026) and Qwen3.7-Plus (June 2026), with Flash arriving last as the efficiency-focused option.

Details

• Native vision-language model: Accepts interleaved text and image input and returns text, as a native multimodal design representing a comprehensive upgrade over the prior Qwen3.6-Flash
• Stronger multimodal foundations: Improved universal object recognition, along with further gains in real-world perception and spatial intelligence
• Upgraded agent capabilities: Significantly enhanced multimodal agent performance for scenarios like Search Agent and CI Agent, with more stable end-to-end task execution
• Optimized multimodal coding: Improved accuracy on multimodal coding tasks
• Context length: Roughly a 1-million-token context window with up to 65,536 tokens of output, enabling long multi-image sequences, long documents, and extended agent trajectories in a single request
• Pricing: Set low at $0.03 per 1M input tokens and $0.13 per 1M output tokens
• Series positioning: The third model in the 3.7 generation, following Qwen3.7-Max (May 2026) and Qwen3.7-Plus (June 2026); its internal snapshot name is “qwen3.7-flash-2026-07-15”

How to try it

• Available via API through Alibaba Cloud’s Model Studio (DashScope), and also offered through third parties such as OpenRouter
• No announcement yet on availability in a consumer-facing chat UI or on any open-weight release; further updates are expected