🧠 Food for thought

Implications, context, and why it matters.

Advanced AI models shift from open-source to proprietary as competitive value increases

• Alibaba’s strategy demonstrates how companies change their open-source approach as models become more powerful and commercially valuable.
• The company’s previous Qwen models achieved massive adoption with over 20 million downloads and 100,000 derivative models on Hugging Face, largely due to their open-source availability1.
• However, both of Alibaba’s “Max” series models—Qwen2.5-Max released in January and now Qwen-3-Max-Preview—are kept proprietary rather than open-sourced1.
• This shift is reflected in pricing, with Qwen-3-Max-Preview costing $0.861 per million input tokens compared to $0.287 for the open-sourced Qwen3-235B model, representing a significant price premium1.
• The pattern suggests that while open-sourcing helps companies build market presence and developer communities, they reserve their most advanced capabilities for direct monetization through API access.

Chinese AI companies rapidly scaling to match US competitors despite significant parameter gaps

• Alibaba’s jump from 235 billion to over 1 trillion parameters represents a more than 4x increase in model size within months of the Qwen3 series launch in May1.
• Despite this rapid scaling, the company still trails significantly behind OpenAI’s GPT-4.5, which is estimated to have 5-7 trillion parameters—roughly 5-7 times larger than Alibaba’s new model1.
• The company’s $52 billion commitment to AI infrastructure over three years exceeds its entire previous decade of investment in this area, indicating the urgency to compete globally1.
• This aggressive investment appears to be generating returns, with Alibaba reporting growth in AI-related products for eight consecutive quarters1.
• The substantial resource commitment reflects the high stakes of the AI race, where parameter scale often correlates with model capabilities and commercial competitiveness.