Neura Market
GLM logo

GLM

Free

A series of open-source MoE language models by Z.ai for agentic tasks. #opensource

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
Company
Z.ai
Visit Website GitHub (40 stars)

About GLM

GLM-5 is a series of open-source Mixture-of-Experts (MoE) language models developed by Z.ai, specifically designed for complex agentic tasks, coding, and long-horizon reasoning. The latest version, GLM-5.2, features a solid 1M-token context, advanced coding capabilities with adjustable thinking effort, and an improved architecture (IndexShare) that reduces per-token FLOPs by 2.9× at long contexts. GLM-5.1 excels at sustained optimization over hundreds of rounds of agentic interaction, while GLM-5 itself scales to 744B parameters (40B active) and integrates DeepSeek Sparse Attention for efficient deployment. The models achieve state-of-the-art open-source performance on benchmarks like Terminal-Bench 2.1, SWE-bench Pro, and NL2Repo, and are freely available on GitHub.

Key Features

Open-source MoE architecture with up to 744B parameters (40B active)
Solid 1M-token context in GLM-5.2 for sustained long-horizon work
State-of-the-art coding performance on Terminal-Bench 2.1 (81.0) and SWE-bench Pro (62.1)
IndexShare sparse attention reduces per-token FLOPs by 2.9× at 1M context
MTP (Multi-Token Prediction) layer improves speculative decoding acceptance length by up to 20%
Multiple thinking effort levels to balance performance and latency
Designed for complex agentic tasks with long-horizon reasoning and iteration
Integration with Z.ai platform for API access

Pros & Cons

Pros
  • Open-source and freely available for research and development
  • Strong performance on coding benchmarks, competitive with closed-source models like Claude Opus
  • Efficient MoE design reduces active parameters while maintaining high capacity
  • Long context window (1M tokens) enables handling of extensive codebases and sessions
  • Capable of sustained optimization over hundreds of rounds without plateauing early
  • Innovative architecture with IndexShare and MTP for improved efficiency and speed
Cons
  • Large total parameter count (744B) may require significant hardware for inference
  • Still slightly behind top closed-source models (e.g., Claude Opus 4.8) on some benchmarks
  • Relatively new model with limited community adoption compared to more established open-source LLMs
  • Documentation and examples primarily focused on GitHub repository, may lack extensive tutorials

Best For

Complex systems engineering and agentic software developmentLong-horizon coding and debugging tasks (Hundreds of rounds of tool calls)Terminal-based automation and real-world command-line tasksRepository generation (NL2Repo) and code commit generationAssisting with ambiguous problem-solving through experimentation and iterationBuilding autonomous coding agents that sustain improvement over time

FAQ

What is GLM-5?
GLM-5 is a series of open-source MoE (Mixture-of-Experts) language models developed by Z.ai, targeting complex systems engineering and long-horizon agentic tasks.
How does GLM-5.2 improve over GLM-5.1?
GLM-5.2 brings a solid 1M-token context, stronger coding capabilities with flexible thinking effort levels, and an improved architecture with IndexShare (reducing FLOPs by 2.9×) and an enhanced MTP layer that increases speculative decoding acceptance length by up to 20%.
What benchmarks does GLM-5 achieve?
GLM-5.2 scores 81.0 on Terminal-Bench 2.1 and 62.1 on SWE-bench Pro, making it the strongest open-source model on standard coding benchmarks. It also leads on NL2Repo and Terminal-Bench 2.0 for the GLM-5.1 version.
Is GLM-5 available for API use?
Yes, GLM-5.2 API services are available on the Z.ai API Platform, and you can try GLM-5.2 at z.ai.
How many parameters does GLM-5 have?
GLM-5 has 744B total parameters with 40B active via MoE, increasing from GLM-4.5's 355B total (32B active). Pre-training data grew from 23T to 28.5T tokens.