GLM-5
GLM-5 targets complex systems engineering and long-horizon agentic tasks.
Verified
Core facts
Read by a script from the official API, not rewritten by a model.
- Repository
- https://github.com/zai-org/GLM-5
- License
- Apache-2.0
- Maintainer
- zai-org
- Homepage
- https://z.ai/blog/glm-5.3
- Verified
What it is
The project is launching GLM-5, which targets complex systems engineering and long-horizon agentic tasks. The project describes GLM-5.1 as its next-generation flagship model for agentic engineering. The project describes GLM-5.2 as its latest flagship model for long-horizon tasks. GLM-5.3 uses the same base model as GLM-5.2.
Main features
Compared to GLM-4.5, GLM-5 scales from 355B parameters (32B active) to 744B parameters (40B active), and increases pre-training data from 23T to 28.5T tokens.
GLM-5 also integrates DeepSeek Sparse Attention (DSA).
The project proposes IndexShare, which reuses the same indexer across every four sparse attention layers.
The project also improves the MTP layer of GLM-5.2 for speculative decoding.
The project says it developed slime, a novel asynchronous RL infrastructure.
GLM-5.3-Flash starts from a newly trained base model, with its architecture and training recipe redesigned around capability and efficiency.
Getting started
API services for GLM-5.3 and GLM-5.3-Flash are available on the Z.ai API Platform.
GLM-5.3 and GLM-5.3-Flash support controlling the thinking budget through the reasoning_effort parameter, which accepts three levels: low, high, and max.
GLM-5.2 accepts only high and max for reasoning_effort.
In the chat template for GLM-5.3 and GLM-5.3-Flash, clear_thinking defaults to false if not passed.
For deployment on the Ascend NPU platform, inference frameworks such as vLLM-Ascend, xLLM and SGLang are supported.
Who it is for
The GLM-5 series supports fine-tuning with Slime (v0.3.0+) and ms-swift (v4.4.0+). Slime (v0.3.0+) is the reinforcement learning framework used by the GLM team. ms-swift (v4.4.0+) supports SFT, PPO, and GRPO.
The project says GLM-5.1 is built to stay effective on agentic tasks over much longer horizons. The project reports that on standard coding benchmarks GLM-5.2 scores 81.0 vs. 62.0 for GLM-5.1 on Terminal-Bench 2.1 and 62.1 vs. 58.4 on SWE-bench Pro. The project says GLM-5 delivers significant improvement compared to GLM-4.7 across a wide range of academic benchmarks.
The project says that on its internal evaluation suite CC-Bench-V2, GLM-5 significantly outperforms GLM-4.7 across frontend, backend, and long-horizon tasks. Vending Bench 2 requires the model to run a simulated vending machine business over a one-year horizon. The project reports that on Vending Bench 2, GLM-5 finishes with a final account balance of $4,432.
The project says GLM-5.3 has a 50% improvement over GLM-5.2 on its in-house Z.ai Code Bench. The project says GLM-5.3 is state of the art on CyberGym for vulnerability discovery.
Related tools
Related tools listed on this site.
LongCat
Model and API platform. This review reads the official LongCat-2.0 technical article and its linked API product page, covering code understanding/generation, reasoning and interaction purposes; the original chat UI was not reviewed in operation.
讯飞星辰MaaS
The public primary description positions iFLYTEK Xingchen MaaS as a developer-facing model fine-tuning service, covering data processing, model optimization, evaluation and deployment/hosting. This review did not run those console workflows.
MMLU
The original benchmark and code repository for Measuring Massive Multitask Language Understanding, using multidisciplinary tests to analyze model knowledge and understanding. Its paper appeared at ICLR 2021.
FlagEval
Evaluation toolkit and open platform for large models. Its public project covers foundation models, training and fine-tuning algorithms, with language, vision, audio and multimodal evaluation directions.
OpenCompass
Model evaluation platform with language and multimodal leaderboards, benchmark resources, and documentation.
HELM
Stanford CRFM’s open-source holistic model-evaluation framework, unifying datasets, model interfaces and multiple metrics. Its README states that it entered maintenance mode on 2026-06-01.
书生大模型
The official InternLM language-model project. Its current README highlights InternLM3-8B-Instruct, normal response and deep-thinking modes, with links to model cards across releases.
Dataify
AI data-services platform. Its official site describes datasets for training, evaluation and RAG, webpage/search/video data APIs and network resources, primarily supplying data for models and applications.
OpenBMB
OpenBMB’s open-model community entry. Its public page organizes MiniCPM, AgentCPM, VoxCPM, UltraRAG and other models/projects, with community updates.
Gemma
The Gemma repository describes an open-weight model family based on Gemini research and a JAX library for using and fine-tuning Gemma. It states that it is not an official Google product.
腾讯混元大模型
An official Tencent Hunyuan project entry. This source covers Hunyuan-Large MoE models, pretrained and instruction variants, and inference and fine-tuning guidance.
LLaMA
This LLaMA review is limited to the Llama 2 reference-model description in the original official repository’s current README. The repository is explicitly deprecated and points to successor model/tool projects; it is not a current-model or hosted-service catalog.
Sources
This page is written from the official sources below. Open them to check the original text.
- GitHub READMEhttps://github.com/zai-org/GLM-5Fetched
Verified ·
This page is written by AI from the official sources above and checked automatically. This site has not tested it. Found a mistake? Tell us.