GLM-4.5 Open Source: Zhipu Merges Reasoning, Code, and Agent into a Unified Model
Zhipu released GLM-4.5 and GLM-4.5-Air in July 2025, providing both API and open weights.
GLM-4.5 marks Zhipu's shift from general-purpose dialogue models to a unified foundation model for agents, expanding developer reach through open weights, domestic and international APIs, and compatibility with mainstream coding agents.
GLM-4.5 is a 355B total parameter, 32B activated parameter MoE model; the Air version is 106B/12B. Both support thinking and non-thinking modes, 128K context, and native function calling.
Competition among domestic models has moved from single benchmarks to a combination of capabilities, open licensing, inference efficiency, and tool ecosystem, and has begun to directly enter the global coding agent supply chain.
Developers can use the same model to cover both fast response and complex reasoning, but should separately verify quality, latency, cost, and tool behavior for each mode.
Observe real-world deployment after open weight release, agent tool success rates, overseas developer adoption, and API service stability.