GLM-4.5

GLM-4.5 is a flagship model series built on a self-developed Mixture of Experts architecture, designed to integrate reasoning, coding, and agentic decision-making into a unified framework. By moving away from specialized, single-task designs, the model aims to provide a balanced intelligence capable of handling the increasingly complex requirements of modern agentic applications. The architecture supports a 128K context window and is available in two distinct configurations: the flagship model with 355 billion total parameters and the more streamlined Air version with 106 billion total parameters.

The series features a hybrid reasoning design that allows users to toggle between a thinking mode for complex, multi-step problem solving and a non-thinking mode for rapid, instant responses. This flexibility makes the models particularly effective for high-volume agent deployments and function-calling pipelines where efficiency is critical. By utilizing a Mixture of Experts approach, the models maintain high performance while optimizing active parameter usage—12 billion for the Air version and 32 billion for the flagship—enabling a practical balance between computational cost and task-specific capability.

ModelScopeZhipuAI/GLM-4.5glm

Open provider page Provider docs

Quick Info

Provider: ModelScope
Model key: ZhipuAI/GLM-4.5
Release date: Jul 28, 2025
Last updated: Jul 28, 2025
Knowledge cutoff: 2025-04
Input modalities
Output modalities
Capabilities

Cost

A provider subscription or plan supersedes token-based pricing for this model.

Limits

Output tokens: 98,304 tokens
Context window: 131,072 tokens

Latest news about GLM-4.5

Zhipu AI

CoverageBenchmark

The llm.ing dashboard provides a current snapshot of GLM-4.5 benchmarks and operational stats as of late September 2026. It lists an Artificial Analysis Intelligence Index of 12.8, a Design Arena 3d preference score of 1194 (with per-category breakdowns dated September 23–26, 2026), and an LMArena Text preference score of 1430 recorded September 13, 2026. These figures indicate GLM-4.5 remains actively benchmarked and ranked across multiple evaluation arenas. The same llm.ing page lists live hosting endpoints for GLM-4.5 via OpenRouter, including a Z.AI fp8 deployment offering a 131K context window at $0.6 per million input tokens and $2.2 per million output tokens, with 99.99% 24-hour uptime reported. Although this is aggregator/mirrored data rather than a first-party Zhipu announcement, it offers a developer-facing reference for current GLM-4.5 availability, pricing, and arena standing.

Zhipu AI

Coverage

According to Pandaily's July 28, 2025 launch report, Zhipu AI (now rebranded as Z.ai) open-sourced GLM-4.5, a 355-billion-parameter flagship foundation model positioned as a native "agent" model for autonomous tasks. The article explicitly describes GLM-4.5 as the first SOTA-level native agent model in China, fusing reasoning, code generation, and interactive decision-making in a single system. It also notes Zhipu's rebrand to Z.ai alongside the launch. The Pandaily piece details the technical profile of GLM-4.5: a Mixture-of-Experts architecture with 355B total parameters and roughly 32B active per query, plus a lighter GLM-4.5-Air variant (106B total / 12B active). Training used 15 trillion tokens of general pre-training followed by 8 trillion tokens of code, reasoning, and agent fine-tuning, with a 128,000-token context window. While the source is a third-party press recap rather than an official Zhipu announcement, it remains the most substantive GLM-4.5-specific evidence in the candidate set.