Alibaba released its most powerful AI model, Qwen 3.8-Max, on August 3, 2026, featuring 2.4 trillion parameters [1, 2]. The model uses a Mixture-of-Experts (MoE) architecture that activates about 95 billion parameters per run to reduce computational costs and latency [1, 2]. It can process up to 1 million tokens at a time and supports multi-modal inputs including text, images, and video [1, 2]. Alibaba plans to officially launch Qwen 3.8-Max on its Alibaba Cloud Model Studio platform the week following August 3 [1, 2].
Qwen 3.8-Max ranked as the top Chinese language model on the Arena.AI text model leaderboard but remains behind Anthropic's Claude Fable 5 and Opus models in overall ranking [1, 2]. Alibaba and other Chinese AI firms emphasize releasing "open-weight" models that developers can download and customize, contrasting with closed models from companies like OpenAI, Anthropic, and Google [1, 2].
Separately, DeepSeek launched its low-cost AI model, V4-Flash, on July 31, 2026, continuing its strategy of ultra-low-cost AI alternatives [3, 1, 2]. Research firm Artificial Analysis found V4-Flash is the cheapest to run among well-known global AI models, costing just $0.14 per million input tokens and $0.28 per million output tokens, with an average benchmark cost of 3 US cents [3, 1, 2]. This compares with $0.86 for Moonshot AI’s Kimi K3, $1.86 for OpenAI GPT-5.6 Sol, and $3.15 for Anthropic Claude Fable 5 [3, 1, 2]. DeepSeek is reportedly preparing for a potential initial public offering [3, 1, 2].
Omdia chief analyst Lian Jye Su said Chinese AI firms have found an important market with models balancing performance, cost, transparency and accessibility. Su noted many business applications do not require the very top-tier models but need ones that are affordable and open-weight, which fits the Chinese model makers' strategy [1].
