FAMILY / Z.AI

GLM

A family that began with GLM-130B and ChatGLM before moving toward reasoning, coding, and long-horizon agent engineering.

First model: 2022
Models: 15
Latest dated version: GLM-Legal

01 / RELEASE SEQUENCE

Family release sequence

Compare versions →

02 / GENERATIONS

What changed

GLM-5.3-FlashX

FlashX separates a serving-speed SKU from the underlying capability band, keeping inference-service changes distinct from base-model weights.

GLM-130B

GLM-130B marked the GLM family’s move into large publicly released models, well before GLM-4.5.

2022-08-04

ChatGLM-6B

ChatGLM-6B brought the GLM approach to a more deployable open-model size and became a direct historical reference for later open GLM-4 models.

2023-03-14

ChatGLM2-6B

The second-generation bilingual dialogue model, using multi-query attention and expanding context to 32K.

2023-06-25

ChatGLM3-6B

The third ChatGLM dialogue generation with stronger tool-use and code-execution support; this is the standard 8K chat model, not the separate 32K/128K versions.

2023-10-27

GLM-4

GLM-4 moved the family from ChatGLM’s open dialogue line into a foundation-model generation serving APIs and multiple products.

2024-01

GLM-4.5

GLM-4.5 made the “native agent model” an explicit family direction while releasing both base and chat weights.

2025-07-28

GLM-4.6

The open model between GLM-4.5 and GLM-4.7, expanding context to 200K and improving coding and agent capabilities.

2025-09-30

GLM-4.7

The open successor to GLM-4.6 with continued gains in coding, reasoning, and agent tasks.

2025-12-22

GLM-5

GLM-5 shifted long-horizon agent evaluation from single code generations toward complete engineering tasks.

2026-02-12

GLM-5.1

A GLM-5 update for long-horizon work, designed to execute autonomously for hours within one task.

2026-04-07

GLM-5.2

Extended context to 1M and optimized for large codebases, long documents, and project-scale agent work.

2026-06-16

GLM-5.3

GLM-5.3 is a clear case of major agent gains coming from scaled post-training, bringing the flagship GLM-5 specification into the open-weight line.

2026-08-14

GLM-5.3-Flash

Z.ai further disclosed a 30T-token multimodal pretraining corpus and Manifold-Constrained Hyper-Connections (mHC), reporting about 3× less attention compute and a 4.4× smaller KV cache than GLM-5.3.

2026-09-15

GLM-Legal is a specialized GLM branch for legal workflows, with domain training and API beta applications confirmed by the originator. The original Z.ai announcement is dated 2026-10-12; the next-day AutoClaw reprint provides the full training details. Parameter count, a distinct API model ID, weights, and license are not disclosed.

2026-10-12