Z.AI Adds GLM-5.3-Flash to Coding Plan but Leaves FlashX Off It

Z.AI’s GLM-5.3-Flash release gives coding-plan subscribers a larger included allowance, but its faster FlashX sibling is not part of that plan. Both models are available through Z.AI’s API, making access terms—not just model speed—the immediate difference between the variants.

Z.AI describes Flash as the first native multimodal model in its GLM-5 series. It accepts video, images, text and files, produces text, and has a one-million-token context window. The model has 320 billion total parameters, with 18 billion activated parameters.

For developers calling the models directly, Z.AI lists the identifiers glm-5.3-flash and glm-5.3-flashx. Vercel AI Gateway also lists FlashX under zai/glm-5.3-flashx, with a one-million-token context window and a maximum output limit of 131,072 tokens.

An efficiency pitch for long inputs

Z.AI says Flash combines sparse and linear attention, a design it positions as a way to reduce the work of handling long context. Compared with GLM-5.3, the company reports a 3.01-fold reduction in attention computation and a 4.44-fold reduction in key-value-cache usage.

The workflows Z.AI is targeting

• Visual coding and interaction across browsers and graphical interfaces.
• Office-document generation, document processing and financial research workflows.
• Video understanding, Blender scenes, game development and CAD-model generation.

The documentation also lists function calling, structured JSON output, streaming responses and context caching. Its thinking mode is enabled-only, so developers cannot turn that mode off. The release’s practical appeal will depend on whether the plan-based Flash option or FlashX’s stated response speed better fits a given workflow.