AMAImedia

Released and maintained by AMAImedia.

• Founder: Ilia Bolotnikov
• Organization: AMAImedia.com
• X (Twitter): @AMAImediacom
• LinkedIn: Ilia Bolotnikov
• Telegram: @djbionicl
• Release date: 2026-08-27

---

GLM-5.3-Flash

Join our WeChat or Discord community.
Check out the GLM-5.3-Flash blog and GLM-5 technical report.
Use GLM-5.3-Flash API services on the Z.ai API Platform.

Introduction

We introduce GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series. With 320B total parameters and just 18B active parameters, it is designed for efficient long-context, coding, agentic, and multimodal workloads.

GLM-5.3-Flash starts from a newly trained base model, with its architecture and training recipe redesigned around capability and efficiency. It combines sparse and linear attention with Manifold-Constrained Hyper-Connections (mHC) and a multimodal pre-training corpus.

Serve GLM-5.3-Flash Locally

GLM-5.3-Flash supports deployment with the following frameworks:

• SGLang — see the cookbook
• vLLM — see the recipes
• TokenSpeed — see the model recipe
• KTransformers — see the tutorial

Usage

Transformers: text generation

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "AMAImedia/GLM-5.3-Flash"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)

messages = [{"role": "user", "content": "Explain the main benefits of multimodal agents."}]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
).to(model.device)

with torch.inference_mode():
    output = model.generate(
        inputs,
        max_new_tokens=512,
        temperature=0.7,
        top_p=0.9,
        do_sample=True,
    )

print(tokenizer.decode(output[0][inputs.shape[-1]:], skip_special_tokens=True))

Transformers: multimodal input

For image questions, use the official GLM processor and multimodal message format documented by the model implementation. Keep image inputs in the message content and use the processor to build model inputs; verify the exact API against the installed Transformers version.

from transformers import AutoProcessor, AutoModelForMultimodalLM

model_id = "AMAImedia/GLM-5.3-Flash"
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForMultimodalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto",
    trust_remote_code=True,
)

messages = [{
    "role": "user",
    "content": [
        {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
        {"type": "text", "text": "What is shown in this image?"},
    ],
}]

inputs = processor.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=128)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))

vLLM

pip install -U vllm
vllm serve AMAImedia/GLM-5.3-Flash \
  --trust-remote-code \
  --tensor-parallel-size 8 \
  --max-model-len 131072

The server exposes an OpenAI-compatible endpoint:

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
response = client.chat.completions.create(
    model="AMAImedia/GLM-5.3-Flash",
    messages=[{"role": "user", "content": "Write a concise summary of this document."}],
    temperature=0.7,
    max_tokens=512,
)
print(response.choices[0].message.content)

Practical deployment notes

Use bfloat16 where supported, or a framework-supported quantized checkpoint when GPU memory is limited. For multimodal serving, follow the selected framework's documented image-input format. Confirm the installed framework version, GPU memory requirements, tensor-parallel configuration, and supported context length before production deployment.

Footnotes

• HLE w/ tools (full set): Evaluation uses temperature=1.0 and top_p=0.95 with a maximum generation length of 163,840 tokens and a maximum context length of 300,000 tokens.
• NL2Repo: Evaluation uses temperature=1.0, top_p=1.0, and max_new_tokens=64k under 1M context.
• DeepSWE: Evaluation uses temperature=0.95, top_p=1.0, timeout=6h, and 400K context.
• Terminal-Bench 2.1: Evaluation uses temperature=1.0, top_p=1, max_new_tokens=65536, and a 6h timeout.
• Toolathlon Verified: Results are reported as pass@1 averaged over three independent runs.
• AutomationBench: Evaluation uses AutomationBench v1.0.6.
• GDPval-AA v2: Models are evaluated by Artificial Analysis.
• BabyVision: Evaluation uses temperature=1.0, top_p=0.95, and a maximum context length of 164K tokens.

Citation

If you find GLM-5.3-Flash useful in your research, please cite the official technical report:

@misc{glm5team2026glm5vibecodingagentic,
  title={GLM-5: from Vibe Coding to Agentic Engineering},
  author={GLM-5-Team},
  year={2026},
  eprint={2602.15763},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2602.15763},
}