88Z AI, 9 September — Ant Bailling releases its first native multimodal large model, Ling-3.0-flash-VL, with 124B total parameters and 5.5B activated per inference. It supports image, text, video, and 256K context, emphasizing low latency and cost.
At a glance
- Ling-3.0-flash-VL has 124B total parameters with only 5.5B activated per inference and supports 256K context.
- Native joint training via visual feedback loops suits low-latency scenarios like web replication and GUI agents.
The update
Ant Group's Bailling large model has officially open-sourced its first native multimodal large model, Ling-3.0-flash-VL. The model features a total of 124 billion parameters but activates only 5.5 billion during single-step inference, significantly reducing computational resource consumption. It possesses multimodal capabilities for processing images, text, and video, while supporting a context window of up to 256K tokens. By employing a visual feedback loop involving observation, action, verification, and correction, the model uses a native joint training strategy that is claimed to enhance pure text processing abilities. Its design prioritizes low latency and low-cost operation, making it suitable for various practical application scenarios such as web page replication, GUI agents, and medical report interpretation.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 百灵大模型Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.