Latest Content
Qwen3 30B A3B Thinking 2507 icon

Qwen3 30B A3B Thinking 2507

NVIDIA
The Qwen3-30B-A3B-Thinking-2507 is a reasoning-focused Mixture-of-Experts (MoE) Causal Language Model designed for high-complexity analytical workloads that require explicit, step-by-step deliberation. This model operates exclusively in thinking mode and is trained to expose its internal reasoning process during generation, making it well-suited for advanced tasks in mathematics, science, coding, and academic problem solving. It supports a native 262,144-token context window, enabling deep reasoning over long documents, large codebases, and multi-source inputs without losing coherence.
TypeMoE LLM
CapabilitiesText Generation, Instruction Following, Reasoning, Mathematical Reasoning+5 more
Release DateApril 28, 2025
Links
Paper|Blog|HF Model Card
LicenseApache 2.0

Inference Instructions

Deploy and run this model on NVIDIA B200 GPUs using the command below. Copy the command to get started with inference.

CONSOLE
docker run -it --rm 
 --runtime=nvidia 
 --gpus all 
 --ipc=host 
 --shm-size=128g 
 -p 8000:8000 
 -e VLLM_USE_DEEP_GEMM=1 
 -v ~/.cache/huggingface:/root/.cache/huggingface 
 -e HF_TOKEN='YOUR_HF_TOKEN' 
 -e VLLM_USE_FLASHINFER_MOE_FP16=1 
 -e LD_LIBRARY_PATH='/usr/local/nvidia/lib64:/usr/local/nvidia/lib:/usr/lib/x86_64-linux-gnu' 
 vllm/vllm-openai:v0.15.0-cu130 
 Qwen/Qwen3-30B-A3B-Thinking-2507 
  --tensor-parallel-size 8 
 --enable-expert-parallel 
 --max-model-len auto 
  --max-num-batched-tokens 65536 
  --gpu-memory-utilization 0.95 
  --max-num-seqs 1024 
 --disable-log-requests 
 --trust-remote-code

Model Benchmarks

Each model was tested with a fixed input size and total token volume while increasing concurrency to measure serving performance under load.

ITL vs Concurrency

Time to First Token

Throughput Scaling

Total Tokens/sec vs Avg TTFT

Vultr Cloud GPU

NVIDIA HGX B200

Deploy NVIDIA B200 on Vultr Cloud GPU

Deploy NVIDIA B200 instancesContact Sales