GLM-5.2 Honey Ultra

GLM-5.2 Honey Ultra sits inside the glm family as a specialized output-compression variant of glm-5.2. According to the supplied listing, it carries a built-in ruleset that compresses both generated code and prose toward an answer-only style, while sharing the same upstream model and per-token price as its sibling glm-5.2. That positioning makes it useful for workloads where brevity and signal density matter more than verbose explanations, such as agent loops, tool-call orchestration, and tightly scoped code generation where every output token is paid for.

In practice, the model's distinguishing trait is its aggressive tier behavior: it produces fewer output tokens than standard glm-5.2 while inheriting its capabilities for reasoning, tool use, structured output, and temperature control, and its open weights allow self-hosting or custom gateway deployment. A 1,000,000-token context window paired with a 131,072-token output limit supports long-running agent and retrieval pipelines, and the same per-token pricing as glm-5.2 keeps the trade-off purely about output efficiency rather than cost.