GLM 4.7 Flash
GLM-4.7-Flash is a text-generation model with a 30B-A3B Mixture-of-Experts (MoE) architecture from zai-org, released under the MIT license. It is designed for lightweight deployment while balancing performance and efficiency. The model supports English and Chinese, integrates with Hugging Face Transformers, and can be served locally using vLLM or SGLang. It is the strongest model in the 30B class according to the official model card, with benchmark results provided for tasks such as AIME 25, GPQA, LCB v6, HLE, SWE-bench Verified, τ²-Bench, and BrowseComp.
At a glance
- Parameters
- 31.22B
- Architecture
- glm4_moe_lite
- Context length
- 202,752
- License
- mit
- Disk space
- 58.2 GB
- Software
- vLLM, Transformers, vLLM
Will it run on your machine?
Save your machine to see a personalized rating and its reasoning.
Add my hardwareBest for
- Text generation in English and Chinese
- Lightweight local deployment
- Multi-turn agentic tasks (e.g., τ²-Bench, Terminal Bench 2) with Preserved Thinking mode
- Code generation and software engineering tasks (e.g., SWE-bench Verified)
- Question answering and reasoning benchmarks (e.g., AIME 25, GPQA, HLE)
Tradeoffs
- May not be suitable for tasks requiring modalities beyond text generation
- Performance on specific benchmarks does not guarantee real-world application results
- Optimal settings may vary by task; default parameters are suggested for most tasks
- Local deployment requires specific inference frameworks (vLLM, SGLang) with main branch support
- No information on hardware requirements, compatibility, or licensing beyond MIT is provided
Ways to run it
vLLM · Linux
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
vllm serve zai-org/GLM-4.7-Flash --tensor-parallel-size 4 --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-4.7-flashTransformers · See official guide
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
MODEL_PATH = "zai-org/GLM-4.7-Flash"
messages = [{"role": "user", "content": "hello"}]
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
inputs = tokenizer.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt",
)
model = AutoModelForCausalLM.from_pretrained(
pretrained_model_name_or_path=MODEL_PATH,
torch_dtype=torch.bfloat16,
device_map="auto",
)
inputs = inputs.to(model.device)
generated_ids = model.generate(**inputs, max_new_tokens=128, do_sample=False)
output_text = tokenizer.decode(generated_ids[0][inputs.input_ids.shape[1]:])
print(output_text)vLLM · Linux
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
vllm serve zai-org/GLM-4.7-FP8 --tensor-parallel-size 4 --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-4.7-fp8Explore its uses
Keep exploring
Bonsai 2 · 27B
A compressed 27B-class reasoning model with publisher-provided GGUF packs and a dedicated llama.cpp fork for CUDA, Metal, and CPU.
DeepSeek V4.1 Flash
A multimodal reasoning model with a compressed key-value cache, published as open weights.
Qwen3.8
Explore 4 versions of Qwen3.8. Compare model sizes, packages, hardware requirements, and available setups.