GLM-5-FP8

GLM-5-FP8

<div align="center"> <img src=https://raw.githubusercontent.com/zai-org/GLM-5/refs/heads/main/resources/logo.svg width="15%"/> </div> <p align="center"> 👋 Join our <a href="https://raw.githubusercontent.com/zai-org/GLM-5/refs/heads/main/resources/wechat.png" target="_blank">WeChat</a> or <a href="https://discord.gg/QR7SARHRxK" target="_blank">Discord</a> community. <br> 📖 Check out the GLM-5 <a href="https://z.ai/blog/glm-5" target="_blank">technical blog</a>. <br> 📍 Use GLM-5 API services on <a href="https://docs.z.ai/guides/llm/glm-5">Z.ai API Platform. </a> <br> 👉 One click to <a href="https://chat.z.ai">GLM-5</a>. </p>

Introduction

We are launching GLM-5, targeting complex systems engineering and long-horizon agentic tasks. Scaling is still one of the most important ways to improve the intelligence efficiency of Artificial General Intelligence (AGI). 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), largely reducing deployment cost while preserving long-context capacity.

Reinforcement learning aims to bridge the gap between competence and excellence in pre-trained models. However, deploying it at scale for LLMs is a challenge due to the RL training inefficiency. To this end, we developed slime, a novel asynchronous RL infrastructure that substantially improves training throughput and efficiency, enabling more fine-grained post-training iterations. With advances in both pre-training and post-training, GLM-5 delivers significant improvement compared to GLM-4.7 across a wide range of academic benchmarks and achieves best-in-class performance among all open-source models in the world on reasoning, coding, and agentic tasks, closing the gap with frontier models.

Benchmark

*: refers to their scores of full set.

†: A verified version of Terminal-Bench 2.0 that fixes some ambiguous instructions. See footnote for more evaluation details.

Footnote

• Humanity’s Last Exam (HLE) & other reasoning tasks: We evaluate with a maximum generation length of 131,072 tokens (temperature=1.0, top_p=0.95, max_new_tokens=131072). By default, we report the text-only subset; results marked with * are from the full set. We use GPT-5.2 (medium) as the judge model. For HLE-with-tools, we use a maximum context length of 202,752 tokens.
• SWE-bench & SWE-bench Multilingual: We run the SWE-bench suite with OpenHands using a tailored instruction prompt. Settings: temperature=0.7, top_p=0.95, max_new_tokens=16384, with a 200K context window.
• BrowserComp: Without context management, we retain details from the most recent 5 turns. With context management, we use the same discard-all strategy as DeepSeek-v3.2 and Kimi K2.5.
• Terminal-Bench 2.0 (Terminus 2): We evaluate with the Terminus framework using timeout=2h, temperature=0.7, top_p=1.0, max_new_tokens=8192, with a 128K context window. Resource limits are capped at 16 CPUs and 32 GB RAM.
• Terminal-Bench 2.0 (Claude Code): We evaluate in Claude Code 2.1.14 (think mode, default effort) with temperature=1.0, top_p=0.95, max_new_tokens=65536. We remove wall-clock time limits due to generation speed, while preserving per-task CPU and memory constraints. Scores are averaged over 5 runs. We fix environment issues introduced by Claude Code and also report results on a verified Terminal-Bench 2.0 dataset that resolves ambiguous instructions (see: https://huggingface.co/datasets/zai-org/terminal-bench-2-verified).
• CyberGym: We evaluate in Claude Code 2.1.18 (think mode, no web tools) with (temperature=1.0, top_p=1.0, max_new_tokens=32000) and a 250-minute timeout per task. Results are single-run Pass@1 over 1,507 tasks.
• MCP-Atlas: All models are evaluated in think mode on the 500-task public subset with a 10-minute timeout per task. We use Gemini 3 Pro as the judge model.
• τ²-bench: We add a small prompt adjustment in Retail and Telecom to avoid failures caused by premature user termination. For Airline, we apply the domain fixes proposed in the Claude Opus 4.5 system card.
• Vending Bench 2: Runs are conducted independently by Andon Labs.

Serve GLM-5 Locally

Prepare environment

vLLM, SGLang, KTransformers, and xLLM all support local deployment of GLM-5. A simple deployment guide is provided here.

• vLLM

  Using Docker as:

  docker pull vllm/vllm-openai:nightly 
  

  or using pip:

  pip install -U vllm --pre --index-url https://pypi.org/simple --extra-index-url https://wheels.vllm.ai/nightly
  

  then upgrade transformers:

  pip install git+https://github.com/huggingface/transformers.git
  

• SGLang

  Using Docker as:

  docker pull lmsysorg/sglang:glm5-hopper # For Hopper GPU
  docker pull lmsysorg/sglang:glm5-blackwell # For Blackwell GPU
  

Deploy

• vLLM

  vllm serve zai-org/GLM-5-FP8 \
       --tensor-parallel-size 8 \
       --gpu-memory-utilization 0.85 \
       --speculative-config.method mtp \
       --speculative-config.num_speculative_tokens 1 \
       --tool-call-parser glm47 \
       --reasoning-parser glm45 \
       --enable-auto-tool-choice \
       --served-model-name glm-5-fp8
  

  Check the recipes for more details.

• SGLang

  python3 -m sglang.launch_server \
    --model-path zai-org/GLM-5-FP8 \
    --tp-size 8 \
    --tool-call-parser glm47  \
    --reasoning-parser glm45 \
    --speculative-algorithm EAGLE \
    --speculative-num-steps 3 \
    --speculative-eagle-topk 1 \
    --speculative-num-draft-tokens 4 \
    --mem-fraction-static 0.85 \
    --served-model-name glm-5-fp8
  

  Check the sglang cookbook for more details.

• xLLM and other Ascend NPU

  Please check the deployment guide here.

• KTransformers

  Please check the deployment guide here.

Citation

Our technical report is coming soon.