Z.AI iconZ.AI: GLM 4.5

Model Type

Proprietary model icon

Proprietary Model

API access only

Recommended Use Cases

Text Generation

Try GLM 4.5

GLM-4.5 is Z.AI's first agent-native foundation model, unifying reasoning, coding, and agentic capabilities in a single MoE architecture with hybrid thinking modes.

We are setting a new benchmark with GLM-4.5, demonstrating that cutting-edge performance can be open, efficient, and affordable. — Zhang Peng, CEO of Z.AI

Overview

Released July 28, 2025 alongside Z.AI's international rebrand, GLM-4.5 introduced the company's first Mixture-of-Experts architecture with 355B total parameters (32B active). It features dual-mode inference: "thinking mode" for complex reasoning and "non-thinking mode" for fast responses.

Key Capabilities

  • 355B total / 32B active parameters (MoE)
  • 128K context window
  • Hybrid thinking modes: Toggle reasoning on/off
  • Agent-native design: Planning, tool use, multi-step execution
  • 22T training tokens (including 7T for code/reasoning)
  • Runs on 8× NVIDIA H20 chips

Performance at Launch

GLM-4.5 ranked 3rd overall across 12 industry benchmarks:

BenchmarkScoreNotable Result
SWE-bench Verified64.2%Ahead of Claude 4 Opus, GPT-4.1
Tool-calling90.6%Beat Claude-4-Sonnet (89.5%)
BrowseCompStrongCompetitive with frontier models

When to Use GLM-4.5

Choose GLM-4.5 when you need:

  • Agent-native capabilities with built-in reasoning/action integration
  • Hybrid thinking modes for flexible latency/accuracy tradeoffs
  • Strong coding with tool-calling abilities
  • Cost-effective deployment (13% of DeepSeek's cost at launch)
  • Established model with extensive community testing

Choose GLM-4.6 or GLM-4.7 when you need:

  • 200K context (vs 128K)
  • Enhanced coding performance
  • Preserved Thinking for multi-turn stability

Choose GLM-4.5 Air when you need:

  • Smaller footprint with 106B/12B active parameters
  • Similar capabilities at lower resource requirements

Role in Series

GLM-4.5 marked a new era:

  1. GLM-4 (Jun 2024): Previous generation
  2. GLM-4-32B (Apr 2025): Dense 32B model
  3. GLM-4.5 (Jul 2025): First MoE, agent-native (this model)
  4. GLM-4.5 Air (Jul 2025): Lightweight variant
  5. GLM-4.6+ (Sep 2025+): Subsequent improvements

Links