GLM-5.3-Flash, the new model from Z.ai, is now available on LM Studio’s Bionic platform, adding multimodal input, a 1 million-token context window, and pricing that undercuts the previous GLM-5.2 by a wide margin. The rollout gives users on Mac and Windows a fresh option for agentic AI workflows.
Since its debut in mid-July, Bionic has steadily grown its lineup of models and capabilities. The platform is built for agentic tasks such as coding, research, and working with documents and files, and it can run either local models on your own hardware or cloud models hosted on US-based servers.
How Bionic Handles the New Model
Cloud models served through Bionic operate under a strict zero-data-retention policy, and GLM-5.3-Flash is no exception. According to LM Studio, the model runs from US-based servers with ZDR enabled by default, addressing a common concern for developers cautious about where their data travels.
The company has been quick to broaden Bionic’s roster. Shortly after the platform launched, it added support for Moonshot AI’s Kimi K3. The arrival of GLM-5.3-Flash continues that pace, with LM Studio noting the model is up to 10 times cheaper to run than GLM-5.2.
What GLM-5.3-Flash Brings to the Table
GLM-5.3-Flash is a 320-billion-parameter Mixture-of-Experts model with 18 billion active parameters. It accepts both image and text inputs and offers a 1-million-token context window, making it suited for large documents and extended sessions.
On the benchmarks highlighted by Z.ai, the model scores ahead of GLM-5.2 while generally landing in the same range as frontier models from Anthropic, OpenAI, Google, and DeepSeek. Before its official reveal, the model had already generated attention while being tested anonymously on OpenCode and OpenRouter under the codename “Ox Alpha.”
LM Studio’s announcement arrived just hours after Z.ai formally unveiled the model. The addition means Bionic users can now tap a cheaper, image-capable option for building and running agentic projects. LM Studio says GLM-5.3-Flash is up to 9-10 times more cost-effective to run than its predecessor.
Source
Image: 9to5mac.com
