All Models

Latest flagship model and the most capable open-weights model for coding: +50% over GLM-5.2 on Z.ai Code Bench, open-source SOTA on Terminal-Bench 3.0 and Agents' Last Exam, state of the art on CyberGym (84.5) for vulnerability discovery

First natively multimodal GLM-5 model: 320B total / 18B active parameters, first open frontier model with hybrid sparse-linear attention, vision native to the coding loop, beats GLM-5.2 at one-tenth the price

Previous-generation flagship: 1M lossless context, coding capability open-source SOTA of its generation, supports stable execution of complex long-term tasks — the base that GLM-5.3 post-training builds on

Coding capability aligned with Claude Opus 4.6, significant improvement in long-term tasks, can work independently for up to 8 hours

Programming capability aligned with Claude Opus 4.5, proficient in Agentic long-term planning and execution

Special optimization for long tasks, good continuity in executing complex long-term tasks

Comprehensive upgrade of general dialogue, reasoning, and agent capabilities, programming stronger and more stable, better aesthetics

Lightweight and high-speed, small size with strong capability, suitable for general scenarios such as Chinese writing, translation, role-playing, etc.

Context enhanced to 200K, proficient in advanced coding, complex reasoning, and tool invocation

A balanced performance and cost general model, suitable for dialogue, analysis, and document processing

High-performance and cost-effective lightweight model, with stable performance in inference, coding, and agent tasks

High-performance and cost-effective ultra-fast version, suitable for business scenarios with low latency and high responsiveness

Supports 1M context length, designed for processing long texts and memory-intensive tasks

Fourth-generation flagship model, strong comprehensive capabilities (old version)

Lightweight model, fast speed and low cost (old version)

Lightweight Fast Version, Faster Response Speed (Old Version)

Agent-Specific Model, Optimized for Agent Scenarios (Old Version)

Flash Enhanced High-Speed Version, Fast Inference Speed, Suitable for High-Concurrency Call Scenarios

First Multimodal Agent Base Model, balancing visual understanding, reasoning, and code generation, supporting multimodal inputs of images, videos, and text

Enhanced visual reasoning capability, natively supporting tool calls and long contexts, with more stable frontend code cloning effects

High-performance visual reasoning model, supporting tool invocation and long context

Visual understanding model, supporting input of image and video files

Lightweight visual reasoning model, proficient in understanding complex scenes and multi-step analysis

Old version of image-text video recognition model (old version)

First-generation multimodal visual model (old version)

Lightweight image-text parsing model, balancing high precision and efficiency in document understanding, supporting complex layout parsing

Mobile intelligent assistant framework, supporting natural language to complete App operation tasks, covering the complete set of mobile operation commands

Flagship image generation model, stronger in complex instruction following and knowledge-intensive generation, outstanding in text rendering, especially for Chinese characters

General image generation model, high-quality generation, rich and diverse style expression, more complete image details

Enhanced image generation model, improved image quality, supports more artistic styles

General image generation model, supports multiple resolutions and styles (old version)

High-intelligence flagship video model, with clear image quality, stronger command compliance and physical simulation, supporting generation of first and last frames

High-quality video generation model, with clear image quality, smooth transitions, richer style expression, and can reduce image distortion

High-speed and low-cost video model, with fast generation speed, natural connection of first and last frames, and stronger consistency with multiple reference images

Standard video generation model, supporting multi-resolution output

Voice synthesis model, supporting ultra-human voice generation and emotional expression, providing streaming and non-streaming interfaces

Voice cloning model, capable of quickly generating similar voice in 3 seconds, supporting delicate emotional expression

High-precision speech recognition model, with low character error rate, supporting custom vocabulary and various dialects

Real-time speech dialogue model, supporting Chinese and English speech understanding and generation, adjustable for emotion, tone, and speed

Real-time audio-video model, supporting video calls and long-term dialogue memory, cross-text, audio, and video real-time inference

Enhanced AI search tool, deeply retrieves web information, supports complex queries and multi-round searches

Standard AI search tool, quickly retrieves web information, suitable for general search scenarios

Sogou Engine search enhancement, combined with Sogou search results to provide more accurate information retrieval

Quark Engine search enhancement, combined with Quark search results to provide more comprehensive content retrieval

Code completion model, suitable for code auto-completion and development assistance, improving coding efficiency

Third-generation vector model, used for semantic retrieval enhancement, clustering, topic modeling, and classification, significantly improving the effect

Second-generation vector model, used for semantic retrieval enhancement, clustering, and classification (old version)

Text reordering model, calculates text relevance score, optimizes recall result sorting and matching effect

Animate conversation model, suitable for emotional companionship and virtual character interaction, supporting more natural character expression

Psychological and emotional support model, with professional counseling and emotional guidance capabilities, helping users understand emotions and cope with problems

Open-source 9B parameter model, supports 128K context, can be locally deployed, MIT license

Third-generation open-source 6B dialogue model, the first generation ChatGLM series, MIT license