Skip to main content
Rase AI Intelligence Newsletter

Models & assistants Model or AI service

GLM-4.5 offers reasoning and direct-answer modes

Z.ai released GLM-4.5 and the smaller GLM-4.5-Air as downloadable language models. The original pair was designed for coding, tool calls and tasks that call for different levels of reasoning. [1][2]

Model access
Downloadable GLM-4.5 and smaller Air variants. Z.ai offered both through its hosted API at launch. [1][2]
License
MIT License for the original GLM-4.5 model card’s published weights. [2]

What the model does

The July 28, 2025 release included public model weights and hosted access at launch. Z.ai documented a thinking mode for complex work and a direct-answer mode for quicker responses. [1][2]

Why it matters

Offering both modes lets a team compare fast responses with more deliberate answers on the same model. The smaller Air variant gives developers another way to compare hardware needs. [1][2]

Who can use it

Developers evaluating code assistants or applications that call a tool can test the published models. Teams running the weights should match the checkpoint to available hardware. [1][2]

Key dates

  1. GLM-4.5 and GLM-4.5-Air introduced [1]

Ways to use it

Compare the two modes on a set of ordinary code or information tasks. Test whether a requested action or tool call is correct and validate the returned data before relying on it. [1][2]

Model limitations

The main checkpoint contains 355B total parameters and activates 32B per token. Running the original weights still requires serving capacity for the full model. [1][2]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source Z.ai original release of GLM-4.5 and GLM-4.5-Air
  2. Primary source zai-org/GLM-4.5
  3. Other source GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models

Related profiles

Models & assistants

Hy4

Tencent (Hunyuan)

It is a language model family with documented preview and subsequent variants.

Medium confidenceEvidence checked

Models & assistants

Qwen4-Max

Alibaba (Qwen team, Alibaba Cloud)

It provides model-based assistance for text and related work tasks.

High confidenceEvidence checked

Models & assistants

Gemini 4 Hydrogen

Google DeepMind

It processes text and other supported inputs for tasks such as code assistance and planning.

High confidenceEvidence checked