Z.AI iconZ.AI: GLM 5

Model Type

Proprietary model icon

Proprietary Model

API access only

Recommended Use Cases

Text Generation

Try GLM 5

GLM-5 is Z.AI's flagship foundation model, designed for agentic engineering with state-of-the-art open-weight performance in coding, reasoning, and long-horizon agent tasks.

GLM-5 is the new leading open weights model! It leads the Artificial Analysis Intelligence Index amongst open weights models with a score of 50—the first open weights model to achieve this milestone.

  • Artificial Analysis

Overview

Released February 2026, GLM-5 scales to 744B total parameters (40B active) with 28.5T training tokens--significant increases from GLM-4.7's 355B/32B and 23T tokens. The model integrates DeepSeek Sparse Attention for efficient long-context processing and uses Z.AI's new "Slime" asynchronous RL framework for training.

Key Capabilities

  • 200K context window with 128K output tokens
  • Agentic Engineering: Autonomous long-range planning, backend refactoring, deep debugging
  • Coding: On par with Claude Opus 4.5 in real-world software engineering
  • Reasoning: Strong math and logic performance with tool integration
  • Low hallucination: Lowest hallucination rate among tested models

Performance Highlights

BenchmarkGLM-5Notes
SWE-bench Verified77.8%Highest among open-weight models
Terminal-Bench 2.056.2%Near Claude Opus 4.5 (59.3%)
BrowseComp (w/ context)75.9%Leads open-weight models
τ²-Bench89.7%Near Claude Opus 4.5 (91.6%)
HLE (w/ tools)50.4%Strong agentic search

When to Use GLM-5

Choose GLM-5 when you need:

  • Maximum open-weight capability for complex systems engineering
  • Long-horizon agentic tasks with autonomous planning
  • Production coding workflows rivaling Claude Opus 4.5
  • Self-hosted deployment of frontier-class intelligence
  • Low hallucination rates for knowledge-critical applications

Consider GLM-4.7 when you need:

  • Similar architecture at lower deployment cost
  • Sufficient for most coding workflows
  • Smaller memory footprint

Consider proprietary models when you need:

  • Absolute maximum performance → Claude Opus 4.6
  • Managed API without self-hosting concerns

Role in Series

GLM flagship evolution:

  1. GLM-4.5 (Jul 2025): First MoE architecture, 355B/32B
  2. GLM-4.6 (Sep 2025): Extended context to 200K
  3. GLM-4.7 (Dec 2025): Enhanced coding and thinking modes
  4. GLM-5 (Feb 2026): New architecture, 744B/40B, SOTA open-weight (this model)

Links

GLM 5 | Try That LLM