DeepSeek, Alibaba Open AI Models: Competition Heats Up
Summary
DeepSeek and Alibaba are now allowing users to download and run their flagship AI models on their own infrastructure. This move provides lower costs and more flexible use for users. DeepSeek launched DeepSeek V4 Pro on August 13th. On the same day, Alibaba Group Holding released the Qwen3.8-2.4T-A95B weight, a first step for its Max class model. This allows users to run models on their own devices without needing the developer's API. What's interesting is that this trend is not limited to China. Meta previously released Muse Glimmer with an open weight and plans to open its flagship model, Muse Spark 1.2. The latest generation of open models is also highlighting the capabilities of AI agents, which can perform tasks more independently. For example, DeepSeek V4 Pro scored 62.7 in a programming agent benchmark, a sharp increase from its preview version. Alibaba's Qwen team showed that Qwen3.8-Max could independently write code for about 16 days and run a virtual e-commerce business. The bottom line is that open models offer significant cost advantages. DeepSeek V4 Pro's cost is said to be less than one-seventh of Grok 4.6, despite both being top performers. This could put pressure on closed model providers and their pricing. This matters because it means more powerful AI tools could become more accessible and affordable for everyone.
This is an AI-generated audio summary. Always check the original source for complete reporting.