ทันเอไอ

Z.ai launches GLM-5.3-Flash with vision support, around 10x cheaper than GLM-5.2

The new open model uses a MoE architecture, supports images and video, and offers a 1-million-token context window

Z.ai, the global brand of Zhipu AI, officially launched GLM-5.3-Flash on August 26, 2026, positioning it as a mid-sized model that delivers better performance while costing around 10 times less to use than GLM-5.2.

The model uses a Mixture-of-Experts, or MoE, architecture, which activates only the parts of the model needed for each task. It has 320B total parameters but activates 18B per token, reducing compute requirements. Published benchmark results indicate that GLM-5.3-Flash outperforms GLM-5.2 overall across several tests.

A key upgrade is its multimodal capability, allowing the model to process both images and video without relying on a separate model. This makes it suitable for analyzing screens, visual documents, and website graphics, in addition to general conversation and coding tasks.

GLM-5.3-Flash also has a 1-million-token context window, enabling it to handle large volumes of documents or code in a single session. However, supporting this context length does not mean the model will recall and use every detail with equal accuracy.

Before launch, the model was available for anonymous testing under the name Ox Alpha through OpenRouter and OpenCode. It was also designed from the outset to support domestically produced Chinese AI chips, reflecting efforts to reduce reliance on foreign hardware.

Z.ai released the model under an open-weights MIT license, allowing developers and organizations to study, fine-tune, or deploy it on their own systems subject to the license terms.