Beijing Zhipu Huazhang Technology Co., Ltd. released a series of major updates on January 16, 2025, including the new end-to-end model GLM-Realtime and upgraded versions of GLM-4-Air, GLM-4V-Plus and other models. All All models have been launched on the bigmodel.cn platform. This update covers multiple modalities such as language, voice, image and video, demonstrating Zhipu’s deep accumulation and innovation capabilities in the field of multi-modal large model technology, and specially launched a Flash full-modal free model, aiming to Lower the threshold for large model application and promote the inclusive development of large model technology.
Beijing Zhipu Huazhang Technology Co., Ltd. announced the launch of a series of new models on January 16, 2025, and launched them on bigmodel.cn. Following the launch of "Zhipu Qingyan" in August, the company has made in-depth exploration in the fields of language, speech, image and video understanding and generation, and launched multi-modal models such as GLM-Voice, GLM-4V, CogView, and CogVideoX.
The new end-to-end model GLM-Realtime released this time realizes low-latency video understanding and voice interaction, incorporates a cappella function, and supports up to 2 minutes of memory and function call functions. The company has also simultaneously upgraded the GLM-4-Air and GLM-4V-Plus models, and is committed to providing the industry's strongest performance and cost-effective language model solutions. Zhipu has always been committed to giving back to the society with advanced large model technology, and has specially set up Flash full-mode free models, covering multiple scenarios such as language, text pictures, text videos and image understanding, to help developers easily achieve application innovation.
GLM-Realtime has a 2-minute content memory capability for video calls, and innovatively implements a cappella singing function in voice interaction, allowing large models to sing in conversations. The company integrates Realtime API into smart glasses and companion dolls so that users can experience near real-time interaction with smart assistants. Realtime further supports the Function Call function, which can rely on its own knowledge and capabilities to flexibly call external knowledge and tools to expand to a wider range of business scenarios. GLM-Realtime API has been launched on the open platform bigmodel.cn, and is currently free to call.
GLM-4-Air has been popular with developers for its high cost performance since its launch. This time it has been fully upgraded to GLM-4-Air-0111. By optimizing training data and processes, its performance in some dimensions is close to that of the larger GLM-4- Plus, at the same time, the model price is reduced to 50% of the original price, lowering the threshold for large model application. The visual understanding model GLM-4V-Plus has also been fully upgraded. The new version has significantly improved performance on multiple public lists. It supports variable resolution function, adapts to image input of different sizes, significantly reduces token consumption in small image scenarios, and supports 4K ultra-clear. Lossless recognition of images and extreme aspect ratio images, with video understanding capabilities of up to 2 hours, providing efficient and accurate solutions for long video understanding and analysis.
Zhipu is committed to the inclusiveness of large models. In order to help developers innovate, it has specially set up a Flash series inclusive model API that is free and open to the whole society. As the industry's first all-modal free series of models, developers can call language, multi-modal understanding, and multi-modal generation functions for free. In the near future, the Flash series will be fully upgraded, including language model GLM-4-Flash, image understanding model GLM-4V-Flash, image generation model CogView-3-Flash, and video generation model CogVideoX-Flash.
The model upgrades and new models released by Zhipu Huazhang not only demonstrate its strong technical strength in the field of artificial intelligence, but also reflect its determination to promote the universalization of large model technology, providing developers and users with more convenient and With more powerful AI tools, it is worth looking forward to the emergence of more innovative applications in the future.