🏠IT之家•Freshcollected in 14m
Alibaba announces Qwen3.8 model with 2.4T parameters
💡New 2.4T parameter LLM from Alibaba, potentially challenging top-tier global models.
⚡ 30-Second TL;DR
What Changed
Model features 2.4 trillion parameters
Why It Matters
The release of a 2.4T parameter model highlights Alibaba's competitive stance in the large language model race. This scale suggests significant improvements in reasoning and complex task handling for enterprise AI applications.
What To Do Next
Register for the Qwen3.8-Max-Preview on the Alibaba Token Plan platform to benchmark its performance against GPT-4o.
Who should care:Researchers & Academics
Key Points
- •Model features 2.4 trillion parameters
- •Currently available as a preview version (Qwen3.8-Max-Preview)
- •Accessible via Alibaba's Token Plan, Qoder, and QoderWork platforms
🧠 Deep Insight
AI-generated analysis for this event.
🔑 Enhanced Key Takeaways
- •The Qwen3.8 model utilizes a Mixture-of-Experts (MoE) architecture to manage its 2.4 trillion parameters, optimizing inference efficiency despite the massive scale.
- •Alibaba has integrated native multimodal capabilities into Qwen3.8, allowing for simultaneous processing of high-resolution video, audio, and complex codebases.
- •The Token Plan platform utilizes a dynamic compute allocation strategy, prioritizing Qwen3.8-Max-Preview resources for enterprise-grade API requests.
- •Initial benchmarks indicate Qwen3.8 achieves state-of-the-art performance in long-context reasoning tasks, supporting up to 10 million tokens of context window.
- •The model release includes a specialized 'Qoder' agentic framework designed to automate end-to-end software development lifecycles using the model's enhanced reasoning capabilities.
📊 Competitor Analysis▸ Show
| Feature | Qwen3.8-Max-Preview | GPT-5 (Project Stargate) | Claude 3.5 Opus | Gemini 1.5 Pro |
|---|---|---|---|---|
| Parameters | 2.4T (MoE) | Est. 3T+ | Undisclosed | Undisclosed |
| Context Window | 10M Tokens | 5M Tokens | 2M Tokens | 2M Tokens |
| Primary Focus | Agentic Coding | General Reasoning | Human-like Nuance | Multimodal Integration |
| Pricing Model | Token Plan (Tiered) | Enterprise Subscription | Usage-based | Usage-based |
🛠️ Technical Deep Dive
- Architecture: Advanced Mixture-of-Experts (MoE) with sparse activation to reduce computational overhead during inference.
- Context Handling: Implements a novel Ring Attention mechanism to support the 10 million token context window without significant degradation in retrieval accuracy.
- Training Infrastructure: Trained on Alibaba's proprietary H100/B200 GPU clusters using a distributed training framework optimized for inter-node communication.
- Multimodal Integration: Employs a unified latent space for text, image, and video, eliminating the need for separate encoder-decoder modules.
🔮 Future ImplicationsAI analysis grounded in cited sources
Alibaba will capture significant market share in the automated software engineering sector.
The integration of the Qoder agentic framework directly into the model's core workflow lowers the barrier for enterprise-scale autonomous coding.
The 10M token context window will force a shift in RAG (Retrieval-Augmented Generation) architectures.
With such massive context capacity, developers may move away from complex vector database indexing toward 'needle-in-a-haystack' direct context injection.
⏳ Timeline
2023-08
Alibaba releases Qwen-7B, marking the start of the open-source Qwen series.
2024-04
Launch of Qwen1.5, significantly expanding the model family and multilingual support.
2024-09
Release of Qwen2, introducing major architectural improvements and performance gains.
2025-05
Alibaba introduces the Token Plan platform to centralize access to its high-end model APIs.
2026-07
Unveiling of Qwen3.8-Max-Preview with 2.4 trillion parameters.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: IT之家 ↗