GLM-5.3-Flash is Z.ai's first natively multimodal model in the GLM-5 series. It is a 320B Mixture-of-Experts model with 18B active parameters, built for coding, agentic work, and visual tasks, and released under the MIT License.

Use GLM-5.3-Flash in LM Studio Bionic

GLM-5.3-Flash is available as a Cloud model in LM Studio Bionic, with up to 1M tokens of context in the current LM Studio Cloud configuration. It accepts text and images and supports tool calling for coding and agentic workflows.

Efficient multimodal architecture

GLM-5.3-Flash uses a newly trained base model built around a hybrid of sparse and linear attention. Linear attention models local dependencies while sparse attention retrieves relevant long-range context. Z.ai also introduces IndexPool, which compresses indexer key vectors to reduce the cost of serving long contexts.

The model uses Manifold-Constrained Hyper-Connections (mHC) and was pretrained on a 30T-token multimodal corpus. Z.ai reports that, compared with GLM-5.3, GLM-5.3-Flash reduces attention computation by 3.0× and KV-cache size by 4.4× while supporting contexts up to 1M tokens.

Visual coding and agentic work

Native visual understanding brings screenshots, interfaces, documents, charts, and other visual artifacts into the model's coding loop. GLM-5.3-Flash can inspect rendered results, work with browser and computer tools, and iteratively refine its output. Z.ai highlights workflows spanning frontend development, office documents, financial analysis, video editing, and 3D scene creation.

The model uses forced thinking with three reasoning effort levels: low, high, and max. Max is the recommended setting for complex coding and long-horizon tasks.

Benchmark results

Z.ai reports that GLM-5.3-Flash improves substantially over GLM-5.2 across coding, agentic, and visual evaluations:

Benchmark	GLM-5.3-Flash	GLM-5.2	DeepSeek-V4-Vision-Exp	Claude Opus 4.8
Terminal-Bench 2.1	84.3	81.0	83.9	85.0
DeepSWE v1.1	63.4	46.2	59.3	58.0
Toolathlon Verified	78.4	59.9	75.9	76.2
AutomationBench v1.0.6	48.8	26.2	38.8	41.0
Agents' Last Exam	26.3	20.4	27.3	27.0
OfficeQA Pro	62.4	—	57.9	48.9
BabyVision	53.4	—	35.1	46.8

Benchmark image and results are from Z.ai's GLM-5.3-Flash launch post. Evaluation settings vary by benchmark; see the launch post and official model card for the complete results and methodology.

Sources and license

GLM-5.3-Flash is released under the MIT License. See Z.ai's launch post, model documentation, and official model card for more details.