On September 22, 2025, Baidu Intelligent Cloud Qianfan unveiled its latest visual comprehension model, Qianfan-VL, and announced its complete open-source availability. Starting immediately and running through October 10, users have the opportunity to explore the 8B and 70B versions of the model on the platform, free of charge. The Qianfan-VL series comes in three distinct variants: 3B, 8B, and 70B. This visual comprehension large model has undergone in-depth optimization to excel in enterprise-level multimodal application environments.

Built upon open-source foundations, the model carries out its entire computational workflow on Baidu’s proprietary Kunlun Core P800 chipset. It is capable of supporting parallel computing across up to 5,000 cards for a single task. The model showcases state-of-the-art (SOTA) performance in both general and specialized task evaluations. With its range of sizes, the Qianfan-VL series is designed to meet the diverse needs of various scenarios.

The 8B and 70B models, in particular, are equipped with chain-of-thought capabilities, making them suitable for complex tasks such as interpreting intricate charts, engaging in visual reasoning, and solving mathematical problems. These models excel in OCR (Optical Character Recognition) and document comprehension, accurately identifying handwritten text, mathematical formulas, and text in natural settings. They are also adept at parsing tables and charts, enabling intelligent document Q&A sessions and structured analysis.