Two Days After Moonshot AI's 'Kimi K3,' Alibaba Introduces a 2.4T-Parameter AI Model
Open Model…Differentiates from Closed U.S. Models
As Chinese AI startup Moonshot AI recently sent shockwaves through the market with its 2.8T-parameter 'Kimi K3', Alibaba countered by unveiling its latest open-weight model 'Qwen 3.8', boasting 2.4T parameters. Chinese AI companies are now challenging the U.S.-dominated frontier AI market by openly releasing models that rival leading AI systems like ChatGPT and Claude in performance.
Alibaba's AI model development unit, Tongyi Qianwen, announced on the 19th (local time) via social media platform X (formerly Twitter), “Qwen 3.8 will soon be released and provided as an open-weight version.”
Alibaba claimed, “With 2.4T parameters, this model is one of the most powerful currently available, ranking second only to Anthropic's Claude 3.5 Sonnet.”
Qwen 3.8 is an ultra-large AI model with 2.4 trillion parameters, significantly expanding the scale from 'Qwen3 Max', which was released in September last year. Parameters are internal variables adjusted during AI training and serve as a key indicator of a model's learning capacity and performance. However, detailed benchmark results for Qwen 3.8 have not yet been disclosed.
The industry analyzes that this announcement, coming just two days after MoonShot AI's unveiling of 'Kimi K3,' signals the intensification of competition among Chinese AI companies. Chinese firms have recently been releasing models that approach the top U.S. levels in both parameter scale and performance, positioning themselves in the race for technological leadership.
Notably, these frontier-level models are being released via an open-weight approach, sharing fully trained weights for anyone to download, run, or fine-tune--distinguishing themselves from the closed AI ecosystems dominated by U.S. companies.
Kimi K3, featuring a 2.8-trillion-parameter Mixture of Experts (MoE) architecture, is characterized by significantly enhanced coding, reasoning, and long-text processing capabilities. MoonShot AI claims that Kimi K3 is a frontier model competitive with Anthropic's Claude 3.5 Sonnet, GPT series, and others. Foreign media have analyzed that this release could mark another “DeepSeek Moment,” significantly narrowing the gap between Chinese AI and U.S. frontier models.
Earlier, DeepSeek unveiled a preview of its 1.6-trillion-parameter 'DeepSeek V4 Pro' in April, and Z.ai introduced the 753-billion-parameter open-weight model 'GLM-5.2' in June, which emphasizes AI agent, coding, and long-term task performance.
Gary Yu, a Morgan Stanley analyst, stated, “Kimi K3 has received positive global evaluations, demonstrating that Chinese LLMs are catching up to U.S. leaders in model scale, performance, and cost.” He added, “We expect more Chinese LLMs with larger scale, higher affordability, and superior performance to emerge as global competitors.”
