Qwen3.8 is launching and going open-weight soon
Alibaba's Qwen3.8, a colossal 2.4T parameter AI model, is launching soon and going open-weight, igniting the fierce competition in the frontier LLM space. This move is welcomed on Hacker News as it promises to further democratize powerful AI, offering an open alternative to proprietary models like Anthropic's Fable 5, which Qwen claims to trail by just a hair. The community eagerly anticipates its release while debating its real-world performance against rivals and hoping for locally runnable versions.
The Lowdown
Alibaba's Qwen AI team has announced the imminent launch of Qwen3.8, a formidable large language model boasting a staggering 2.4 trillion parameters. Breaking ranks with many closed-source frontier models, Qwen3.8 is set to be released as open-weight, making it accessible to a broader community of developers and researchers. The announcement touts Qwen3.8 as one of the most powerful models available today, claiming performance second only to Anthropic's Fable 5.
- Massive Scale: Qwen3.8 features 2.4 trillion parameters, positioning it as a contender in the very top tier of LLMs.
- Open-Weight Release: The decision to release the model as open-weight allows for greater transparency, customization, and local deployment, fostering innovation outside of proprietary ecosystems.
- Competitive Claim: The model is asserted to be compatible with leading frontier AI models, with its performance ranked just below Fable 5, a high bar in the current LLM landscape.
- Early Access: A preview, Qwen3.8-Max-Preview, is already available on Alibaba's Token Plan, Qoder, and QoderWork platforms, inviting developers to test its capabilities.
This open-weight release by Alibaba signals an intensifying arms race in the AI domain, particularly among models seeking to challenge the current leaders. The accessibility of such a large and purportedly powerful model could significantly impact the development trajectory of AI, pushing towards more open and community-driven innovation.
The Gossip
Performance Ponderings & Pareto Pains
The actual performance of Qwen3.8 and its predecessors is a major point of discussion, with users keenly comparing it to other leading models. While some characterize Qwen as a 'benchmark princess' that underperforms in real-world applications compared to models like DeepSeek V4 Pro or Kimi K3, others laud its agentic capabilities, particularly for tasks in Chinese. The bold claim of being 'second only to Fable 5' sparks both skepticism and anticipation, as users weigh the model's token hunger, censorship tendencies, and cost-effectiveness against its perceived intelligence.
Open-Weight Warfare & Geopolitical Gambit
Commenters widely interpret Qwen3.8's open-weight announcement as a strategic maneuver within the competitive and geopolitical landscape of AI. Many believe it's a direct response to Moonshot AI's Kimi K3, fueling an 'open-weight warfare' that ultimately benefits users. The motivation is debated, ranging from Alibaba's intent to commoditize intelligence and drive adoption of its cloud services to China's need for alternatives given restrictions on using US APIs. The broader sentiment champions the 'open vs. closed' model debate, advocating for open-source AI as a safeguard against proprietary lock-in and censorship, citing incidents like HuggingFace's API lockout as justification for accessible, unrestricted models.
Local LLM Longings & Sizeable Solutions
Despite the gargantuan 2.4T parameter count, a recurring desire among users is for smaller, more manageable versions of Qwen3.8 that can be run locally. Many currently appreciate the utility of existing Qwen models (like the 35B MoE and 27B dense) for local tasks, especially when dealing with sensitive data, thus avoiding reliance on external APIs. There's a strong hope for new MoE models in the 35B-122B range that could offer a practical balance of power and efficiency for home computers, making frontier AI more accessible without needing cloud-scale infrastructure.