Beijing Zhiyu Huazhang Technology Co., Ltd. announced the launch of a series of new models on January 16, 2025, available at bigmodel.cn. Following the release of "Zhiyu Qingyan" in August, the company has deeply explored understanding and generation in the fields of language, speech, image, and video, introducing multimodal models such as GLM-Voice, GLM-4V, CogView, and CogVideoX.
The newly released end-to-end model GLM-Realtime achieves low-latency video understanding and voice interaction, incorporating a singing feature and supporting memory for up to 2 minutes along with Function Call capabilities. The company has also upgraded the GLM-4-Air and GLM-4V-Plus models, aiming to provide the industry’s strongest performance and cost-effective language model solutions. Zhiyu is committed to giving back to society with advanced large model technology, having specially established the Flash series of free multimodal models, covering various scenarios such as language, text-to-image, text-to-video, and image understanding, to help developers easily achieve application innovation.
GLM-Realtime features a 2-minute content memory capability for video calls and innovatively implements a singing function for voice interaction, allowing the large model to sing during conversations. The company has integrated the Realtime API into smart glasses and companion dolls, enabling users to experience near real-time intelligent assistant interactions. Realtime further supports Function Call capabilities, flexibly leveraging external knowledge and tools based on its own knowledge and abilities, expanding into broader commercial scenarios. The GLM-Realtime API is now available on Zhiyu's open platform bigmodel.cn and is currently free to use.
Since its launch, GLM-4-Air has been popular among developers due to its high cost-effectiveness. It has been fully upgraded to GLM-4-Air-0111, optimizing training data and processes, achieving performance on certain dimensions close to the larger GLM-4-Plus, while reducing the model price to 50% of its original cost, thus lowering the barrier to entry for large model applications. The visual understanding model GLM-4V-Plus has also undergone a comprehensive upgrade, with significant improvements in performance on various public leaderboards, supporting variable resolution functions to accommodate different image sizes, significantly reducing token consumption in small image scenarios, and supporting lossless recognition of 4K ultra-high-definition images and extreme aspect ratio images, with a video understanding capability of up to 2 hours, providing efficient and accurate solutions for long video understanding and analysis.
Zhiyu is dedicated to making large models accessible to all, and to support developer innovation, it has specifically set up the Flash series of APIs for free public access. As the industry’s first fully multimodal free series of models, developers can freely access language, multimodal understanding, and multimodal generation functionalities. The Flash series will soon undergo a comprehensive upgrade, including language model GLM-4-Flash, image understanding model GLM-4V-Flash, image generation model CogView-3-Flash, and video generation model CogVideoX-Flash.
