SetupAI
繁
SourceStalecollected in 3h

Alibaba Opens Qwen3.8 Model Weights

Read original on IT之家
#mixture-of-experts#long-context#ai-agents#modelscope

A 2.4T open-weight MoE model with 256K native context is now available for experimentation.

30-Second TL;DR

What Changed

The model contains 2.4T total parameters and activates 95B parameters per token.

Why It Matters

The release gives developers access to a frontier-scale open-weight model that may enable more customization, self-hosting, and domain adaptation. Its extreme context capacity and agent-focused training could benefit complex document and long-horizon workflow applications, though deployment costs and hardware requirements will be substantial.

What To Do Next

Download Qwen3.8-2.4T-A95B from ModelScope and benchmark its long-context inference, memory use, and agent workflows on your target hardware.

Who should care:Developers & AI Engineers

Key Points

  • •The model contains 2.4T total parameters and activates 95B parameters per token.
  • •It supports a native 262,144-token context window, expandable to 1,010,000 tokens.
  • •Each MoE layer has 512 experts; every token selects 10 routed experts plus one shared expert.
  • •Multi-Token Prediction training enables supported inference engines to predict several future tokens.
  • •Weights are available through the ModelScope Qwen3.8-2.4T-A95B repository.

Deep Insight

AI-generated analysis for this event — not the original article.

Enhanced Key Takeaways

  • •The model employs a hybrid attention architecture where 69 of its 92 layers utilize linear attention to manage KV cache memory, while the remaining 23 layers use full attention.
  • •The open-weight release is strictly text-only and mandates a 'thinking' mode, meaning all responses are automatically wrapped in tags, unlike the hosted Qwen3.8-Max version which supports vision input and non-thinking modes.
  • •Alibaba's release includes a built-in Multi-Token Prediction (MTP) draft head, which allows inference engines to speculate and verify multiple future tokens simultaneously to improve generation speed.
  • •The model's 2.4 trillion parameters require significant infrastructure; full-precision (BF16) weights occupy approximately 4.9TB of storage, though highly compressed 1-bit quantization variants (e.g., Unsloth's TQ1_0) can reduce this to under 400GB.
  • •The release is positioned as a direct competitor to other massive MoE models like Moonshot AI's Kimi K3, with Alibaba reporting performance metrics that place it near the top of current benchmarks, including OSWorld-Verified.

Competitor Analysis

Qwen3.8-2.4T-A95B
Total Parameters
2.4T
Active Parameters
95B
Context Window
1M
Key Differentiator
Hybrid linear/full attention; MTP-optimized
Moonshot Kimi K3
Total Parameters
2.8T
Active Parameters
104B
Context Window
1M+
Key Differentiator
High-capacity rival in the Chinese market
DeepSeek V4-Pro
Total Parameters
1.6T
Active Parameters
~40B
Context Window
N/A
Key Differentiator
Pioneer in efficient sparse MoE scaling

Technical Deep Dive

  • Architecture: Sparse Mixture-of-Experts (MoE) with 92 layers.
  • Expert Configuration: 512 total experts; 10 routed experts and 1 shared expert active per token.
  • Attention Mechanism: Hybrid design alternating between full attention and linear attention (69 linear layers, 23 full attention layers).
  • Context Handling: 262,144 tokens native, extensible to 1,010,000 tokens.
  • Inference Optimization: Native support for Multi-Token Prediction (MTP) and compatibility with vLLM, SGLang, and TokenSpeed.
  • Precision: Supports FP8 and BF16 for standard deployment; supports extreme quantization (1-bit/1.5-bit) via custom codebooks.

Future ImplicationsAI analysis grounded in cited sources

Open-weight models will increasingly adopt hybrid linear-attention architectures.
The success of Qwen3.8 in maintaining long-context performance while reducing KV cache constraints demonstrates a viable path for scaling models beyond 1M tokens.
Inference-time 'thinking' will become a standard requirement for frontier-class open models.
By mandating thinking mode in the open-weight release, Alibaba signals that complex reasoning capabilities are now inextricably linked to multi-step chain-of-thought processing.

Timeline

2026-05-18
Release of Qwen3.7-Max and Qwen3.7-Plus.
2026-07-19
Alibaba Cloud introduces Qwen3.8-Max-Preview via token-based subscription.
2026-08-03
General availability of Qwen3.8-Max on Qwen Cloud.
2026-08-12
Official release of Qwen3.8-2.4T-A95B open weights on ModelScope and Hugging Face.
Read original article on IT之家

Weekly AI Recap

Read this week's curated digest of top AI events →

PostLinkedIn

AI-curated news aggregator. All content rights belong to original publishers.
Original source: IT之家 ↗

This is a summary, not the original. Read the source, or get the weekly briefing.

The weekly digest

One email a week. Unsubscribe anytime.