GLM-5.3-Prime

Overview

GLM-5.3-Prime is a language model from Z.ai. It takes text as input and returns text as output, and it is published with support for reasoning and for tool calling.

Z.ai describes this entry as a high-speed variant that uses inference acceleration to deliver higher output throughput. On RealRelay the model is callable under the ID alibaba/glm-5.3-prime through an OpenAI-compatible endpoint.

The context window is 1M tokens, which is the full working space for a single request in this model.

Capabilities

GLM-5.3-Prime supports reasoning and tool calling. Tool calling lets an application expose its own functions to the model so that the model can ask for them during a request.

Input modality is text and output modality is text, so prompts and completions stay in plain text form.

The context window spans 1M tokens, and inference acceleration is what raises output throughput relative to a standard run of the same weights.

Pricing

GLM-5.3-Prime is metered per token. Input is no longer offered per 1M tokens and output is no longer offered per 1M tokens. Cached input reads are billed at no longer offered per 1M tokens.

Every rate above is quoted per 1M tokens, so a request is priced by the tokens it sends and the tokens it returns, with the 1M-token context window setting the upper bound on a single request.

Use cases

Reasoning work that needs a large working space fits the 1M-token context window, since a single request can carry an input of that size.

Applications that combine a reasoning step with calls into their own functions can use tool calling to route those calls.

Workloads that generate a lot of output text can take advantage of the higher output throughput that inference acceleration provides.