GLM-4.7-Flash is an open-source coding model released by Z.ai as the lightweight variant of the GLM-4.7 family, built on a new base model and explicitly specializing in coding and tool calling workloads. According to its official distribution listing, it is a 30B-A3B mixture-of-experts design, meaning roughly 30 billion total parameters with around 3 billion active per token, which Z.ai positions as a small and efficient option that balances performance and resource requirements for lightweight deployment. The model is distributed under the MIT license and made available in GGUF and MLX formats, giving developers flexible options for local inference and quantization on consumer or prosumer hardware.

Practically, GLM-4.7-Flash targets teams that want capable code generation, reasoning, and tool use without paying for the largest proprietary systems. Its MoE structure keeps active compute per token modest, which helps deliver predictable latency and makes quantization straightforward, so it fits well in agentic coding pipelines, IDE assistants, and self-hosted retrieval-augmented setups where stable response times matter. Because it inherits Z.ai's coding-focused training emphasis from the broader GLM-4.7 line and ships with reasoning and tool-calling support out of the box, it is a pragmatic choice for developers who need an open-weight workhorse for software engineering tasks rather than a general-purpose conversational assistant.