Model ID
Qwen/Qwen3.8-2.4T-A95B
Capabilities
Qwen3.8 2.4T A95B
Qwen3.8 2.4T A95B is a frontier-scale Mixture-of-Experts model designed for advanced reasoning, coding, professional work, research, and long-horizon agentic tasks. It features 2.4T total parameters with 95B activated, using a 92-layer hybrid architecture with an 8,192 hidden size and 64 attention heads. The model combines Gated DeltaNet linear attention with Gated Attention, using 512 experts with 10 routed experts and 1 shared expert activated per token. Supporting a native 262K-token context window extensible to approximately 1.01M tokens, it is optimized for complex reasoning, autonomous agent execution, coding, and sustained long-context workloads.
About this model
Qwen3.8 2.4T A95B is a frontier-scale Mixture-of-Experts model designed for advanced reasoning, coding, professional work, research, and long-horizon agentic tasks. It features 2.4T total parameters with 95B activated, using a 92-layer hybrid architecture with an 8,192 hidden size and 64 attention heads. The model combines Gated DeltaNet linear attention with Gated Attention, using 512 experts with 10 routed experts and 1 shared expert activated per token. Supporting a native 262K-token context window extensible to approximately 1.01M tokens, it is optimized for complex reasoning, autonomous agent execution, coding, and sustained long-context workloads.
Model specifications and capabilities are published by the model author and reproduced here from HF Model Card (Qwen/Qwen3.8-2.4T-A95B). Released under Qwen3.8-Max. Further reading: Blog
Deploy Private Inference Endpoint
Choose your hardware and inference engine to get deployment commands and performance benchmarks tailored to your infrastructure
Hardware
B200
1,440 GB
Engine
vLLM
SGLang
Deployment code
Access Restricted
Enter your email to get access to this content
This is a two-node inference setup. To configure the worker node, simply update $IFACE_NAME to match its local network interface, change the rank parameter to --node-rank 1 and add the --headless flag to the command.
Performance under load
Tokens/s (total)
Benchmarks measured by Vultr on B200 · vLLM. Throughput and latency vary with concurrency, input length and engine configuration, so treat these as a comparison baseline, not a service guarantee.
NVIDIA HGX B200
Unlock Next-Generation AI Power on Vultr Cloud GPU