Z.ai released GLM-5.3-Flash, its first native multimodal GLM-5 model, with MIT-licensed weights

Archive entry: this covers an event from August 26, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On August 26, 2026, Z.ai released GLM-5.3-Flash, a 320B-parameter MoE model with 18B activated parameters, native visual input and hybrid linear and sparse attention. The Z.ai release notes described it alongside FlashX and FlashX-Legal variants.

What changed

• Z.ai called it the first native multimodal model in GLM-5.
• Attention computation and KV cache were reduced by 3.01x and 4.44x compared with GLM-5.3.
• Input covered video, image, text and file, and output was text, with 1M context and 128K output.
• The model codes were glm-5.3-flash, glm-5.3-flashx and glm-5.3-flashx-legal, and FlashX ran at 200 tokens/s.
• It targeted office, financial and document workflows, with a FlashX-Legal variant.
• On the GLM Coding Plan it got 3x the quota of GLM-5.3, while FlashX was not on the plan at launch.

Weights and third-party availability

The weights were on the Hugging Face model page as zai-org/GLM-5.3-Flash under the MIT license, and Z.ai documented the model in its GLM-5.3-Flash and FlashX guide. On October 22, 2026, Mistral's platform added the model at $0.15 per million input tokens, $0.03 per million cached tokens and $0.50 per million output tokens, according to the Mistral changelog.

What it means for people who build with AI

Coding Plan users got a cheaper multimodal GLM model with three times the quota, and self-hosters got MIT-licensed weights.

Sources

• New Released (Z.ai (Zhipu), August 26, 2026, official)
• GLM-5.3-Flash/FlashX (Z.ai (Zhipu), official)
• zai-org/GLM-5.3-Flash (Z.ai (Zhipu), official)
• Changelog | Mistral Docs (Mistral AI, October 22, 2026)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.