Sign in
← AI PulseAug 12, 2026

Wire · news · Single-source brief

Alibaba's Qwen3.8-2.4T-A95B Model Deployable on NVIDIA GB300 NVL72

Alibaba has released the open weights for Qwen3.8-2.4T-A95B (Qwen3.8-Max), its largest open-weight model, which NVIDIA is optimizing for multinode deployments on the GB300 NVL72 platform.

By Illumora Editorial

Source · Aug 12, 2026, 6:23 PM · On Illumora · Aug 12, 2026, 6:27 PM

Media from the primary source — shown here so you can stay on Illumora.

Rewritten from one allowlisted primary — not independent enterprise reporting. Lanes →

Brief drafted by Illumora’s editorial model from the linked primary source. Ops desk reviews flagged pieces. How we write →

Read the source →NVIDIA Developer Blog — Serve Qwen3.8-2.4T-A95B, a 2.4T-Parameter Model, with Configurable Reasoning on NVIDIA GB300 NVL72 | NVIDIA Technical Blog
Save

Alibaba has released the open weights for Qwen3.8-2.4T-A95B (Qwen3.8-Max), its largest open-weight model, which features 2.4T total parameters with 95B activated per token. NVIDIA is collaborating with the open-source ecosystem to enable multinode deployments of this model, providing optimized kernels, inference runtimes, and distributed serving recipes.

This model is designed for demanding reasoning and agentic workloads, offering a context window of up to one million tokens and an output length of up to 128K.

Key Points

  • Alibaba released the open weights for Qwen3.8-2.4T-A95B (Qwen3.8-Max), its largest open-weight model.
  • The model has 2.4T total parameters with 95B activated per token and a context window of up to one million tokens.
  • It uses a fine-grained mixture of experts (MoE) architecture with a hybrid of full and linear attention.
  • On Day 0, the model achieves over 4K tokens per second per GPU and over 350 tokens per second per user on NVIDIA GB300 NVL72 in FP8 precision.
  • Built-in reasoning controls (low/high/xhigh) allow developers to configure inference depth per request.
  • The NVIDIA GB300 NVL72 integrates 72 NVIDIA Blackwell Ultra GPUs into a single platform.
  • Developers can deploy the model using a model-free NVIDIA NIM or via SGLang, vLLM, and NVIDIA Dynamo.

Context

According to the NVIDIA Developer Blog, deploying a 2.4T parameter open-weight model requires data-center-scale accelerated compute. Inference at this scale depends on extreme co-design across chips, system architecture, and software. NVIDIA is working to bring the model to multinode deployments.

Why It Matters

This development enables AI factories to run large-parameter models like Qwen3.8-2.4T-A95B in production with high throughput and low latency, addressing complex agentic workloads such as coding and large-scale document analysis.

What To Do

  • Download Qwen3.8-2.4T-A95B model weights from Hugging Face or ModelScope.
  • Deploy the model using a model-free NVIDIA NIM from NVIDIA NGC.
  • Explore NVIDIA NeMo AutoModel for post-training and fine-tuning the model for domain-specific use cases.
  • Test the configurable reasoning controls (low/high/xhigh) to balance compute and reasoning quality for different tasks.