GLM-5.1

Z.ai's April 2026 MIT-licensed 754B agentic-coding model, built to keep improving over hundreds of rounds; GLM-5.2 and 5.3 followed.

In short. GLM-5.1 is a very large text model from Z.ai for coding and long agent tasks. Its license is plain MIT, so companies can use and change it freely. It needs a GPU server or a machine with very large memory, and Z.ai has since released GLM-5.2 and GLM-5.3, so new projects should look at those first.

Summary

Z.ai released GLM-5.1 on 7 April 2026 as its flagship for agentic engineering, two months after GLM-5. Hugging Face counts about 754 billion parameters, the same as GLM-5, whose card gives 744 billion in total and 40 billion active per token. The configuration lists 78 layers, 256 routed experts with 8 used per token plus one shared expert, and 202,752 positions. The model reads and writes text. Z.ai says earlier models, GLM-5 included, apply familiar techniques, gain quickly and then plateau, while GLM-5.1 is meant to keep improving over hundreds of rounds and thousands of tool calls. Its release notes say it can work independently for up to 8 hours in one run, and that it was built with multi-turn SFT, reinforcement learning and a process-quality evaluation framework.

In Z.ai's own table GLM-5.1 scores 58.4 on SWE-Bench Pro (GLM-5 55.1, Claude Opus 4.6 57.3, GPT-5.4 57.7, Gemini 3.1 Pro 54.2), 42.7 on NL2Repo (GLM-5 35.9), 63.5 on Terminal-Bench 2.0 with Terminus-2 (GLM-5 56.2) and 69.0 as its best self-reported run with Claude Code, 68.7 on CyberGym, 68.0 on BrowseComp, 70.6 on τ³-Bench and 71.8 on MCP-Atlas. It still trails the closed models on reasoning, with 31.0 on Humanity's Last Exam without tools against 45.0 for Gemini 3.1 Pro, and 86.2 on GPQA-Diamond against 94.3. All of these scores are self-reported.

The BF16 repository is 1,507.8 GB and an FP8 build of 756.2 GB is published separately. By our estimate a 4-bit build needs about 520 GB of memory. The card lists SGLang 0.5.10 or newer, vLLM 0.19.0 or newer, xLLM, Transformers and KTransformers, and GLM-5's card shows a vLLM command that uses 8-way tensor parallelism. Z.ai replaced it with GLM-5.2 in June and GLM-5.3 in August 2026. In the 30 days to 11 November 2026, Hugging Face counted about 837,000 downloads of the main repository and about 18,600 of the FP8 build, and the main repository's count may include mirrors and automated pulls.

Good for

  • Self-hosted coding agents and long-horizon engineering tasks on a GPU server
  • Teams that want a plain MIT license with no regional or revenue conditions
  • Comparing agent behavior across the GLM-5, 5.1, 5.2 and 5.3 generations
  • Research on large mixture-of-experts serving with sparse attention

Watch out for

  • GLM-5.2 and GLM-5.3 are newer, so this page is kept for reference and for existing deployments
  • It needs a multi-GPU node even in FP8, or a machine with roughly 512 GB of fast memory for a 4-bit build
  • It reads text only, and the context is 202,752 positions, far below the 1M tokens of GLM-5.2
  • All benchmark numbers come from Z.ai and use its own harness settings

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Estimates for the GLM-5.1 (BF16) variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.

Variants Parameters BF16 FP8 / INT8 4-bit (Q4) Fits on
GLM-5.1 (BF16) BF16 754B 40B active 1.73 TB 867 GB 520 GB GPU server
GLM-5.1-FP8 FP8 754B 40B active 1.73 TB 867 GB 520 GB GPU server

Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.

VRAM calculator

The license in plain words

The shortest and simplest permissive license. You can do almost anything with the model as long as you keep the copyright and license notice.

More about MIT

Frequently asked questions

What is GLM-5.1?
GLM-5.1 is a very large text model from Z.ai for coding and long agent tasks. Its license is plain MIT, so companies can use and change it freely. It needs a GPU server or a machine with very large memory, and Z.ai has since released GLM-5.2 and GLM-5.3, so new projects should look at those first.
Can I use GLM-5.1 commercially?
Yes. GLM-5.1 is released under MIT, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
Is GLM-5.1 open source?
GLM-5.1 is released under MIT, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
Can I run GLM-5.1 on my own computer?
Not on a desktop. The smallest variant, GLM-5.1 (BF16), needs about 520 GB at 4-bit precision, which means a GPU server with several data-center cards.
When was GLM-5.1 released?
GLM-5.1 was released by Z.ai (Zhipu AI) on April 7, 2026.
How long a context window does GLM-5.1 have?
GLM-5.1 supports a context window of up to 198K tokens.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 11, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
GLM-5 Z.ai (Zhipu AI) 754B 40B active 198K MIT (Permissive) 520 GB GPU server
GLM-5.2 Z.ai (Zhipu AI) 753B 40B active 1M MIT (Permissive) 520 GB GPU server
GLM-5.3 Z.ai (Zhipu AI) 753B 40B active 1M GLM-5.3 License (Conditional) 520 GB GPU server
DeepSeek-V3.2 DeepSeek 685B 37B active 160K MIT (Permissive) 473 GB Big-memory desktop