A growing number of Chinese artificial intelligence (AI) companies are turning to open-source strategy to promote access to advanced models, with Alibaba's Qwen and Shanghai-based MiniMax unveiling new models this week and announcing plans to open-source them, aiming to reduce costs and remove barriers for developers and users.

Alibaba's Qwen team on Monday released Qwen3.8-Max, the latest and most powerful model in the Qwen family, and said the model's weights would be made publicly available next week. The company also announced plans to open-source the Qwen3.8-27B model, according to AliRearch, the company's WeChat account.

Built on the Qwen 3.5 architecture, Qwen3.8-Max expands its parameter scale to 2.4 trillion, with 95 billion activated parameters and support for a 1 million-token context window. The model delivers comprehensive improvements in coding, office applications, scientific research and long-horizon tasks, enabling it to tackle more complex challenges, complete sophisticated tasks more reliably from end to end, and generate more trustworthy outputs.

The Qwen3.8 API service is now available to developers worldwide through the Qwen AI platform. The service is priced at 12 yuan ($1.7) per million input tokens and 36 yuan per million output tokens in China, while overseas users are charged $2 per million input tokens and $6 per million output tokens, with implicit cache-hit pricing at $0.25 per million tokens. Compared with overseas frontier models with similar intelligence capabilities, Qwen3.8 offers higher cost efficiency, according to the company.

According to the latest rankings released by third-party benchmark platform Arena, Alibaba's Qwen model trailed only Anthropic's Claude series, placing it among the world's top-performing large language models overall.

Meanwhile, Shanghai-based AI company MiniMax released its next-generation multimodal generative model, MiniMax H3, and open-sourced it on Monday, the company told the Global Times in a statement. Sixteen chipmakers and developer communities, including Huawei Ascend, Moore Threads, AMD and Intel, announced support for the model the same day.

According to the company, MiniMax H3 is a general-purpose, omni-modal generative system that ranks first globally for video editing on the Artificial Analysis leaderboard. It can jointly understand multimodal contexts comprising text, images, video and audio, and generate videos of up to 2K resolution and 15 seconds in length with native stereo sound.

Designed to generalize across tasks, H3 acquired broad multimodal understanding and generation capabilities during pre-training, enabling it to follow complex multimodal instructions with a high degree of accuracy, the company said.

Chinese open-source models such as Alibaba's Qwen, DeepSeek and Zhipu AI's GLM have gained growing attention among global developers due to their performance and accessibility.

According to OpenRouter's latest weekly LLM leaderboard, DeepSeek V4 Flash ranked as the most-used model, processing 7.22 trillion tokens. Eight of the top 10 models were developed by Chinese companies, including DeepSeek, Xiaomi, Tencent, Zhipu AI, MiniMax, StepFun and Moonshot AI. Together, they processed about 29.55 trillion tokens, accounting for roughly 87 percent of total usage among the top 10 models on the platform.

Ma Jihua, a veteran telecom analyst, told the Global Times that external restrictions on advanced AI chips have pushed Chinese firms away from the capital-intensive US model of scaling closed systems with massive computing power. Instead, they have prioritized efficiency, lower inference costs and local deployment, using open source to turn these strengths into broader ecosystem advantages.

Ma said closed-source models are essentially products, while open-source models create ecosystems. By releasing model weights and tools, companies can quickly attract developers, build application networks around their models and create stronger long-term advantages than simply selling API access. Chinese open-source models have therefore gained broader application space more quickly than many closed models.

Open source has also significantly reduced costs for smaller companies and independent creators, Ma said. Developers can build on mature models rather than train systems from scratch, while lower API prices shorten deployment cycles and make it easier to use AI in customer service, content production and coding. This has allowed innovation to spread beyond large laboratories into highly specialized real-world scenarios.

Ma said that by open-sourcing its models, China is embedding its technological standards and innovations into global developers' workflows, with Chinese models increasingly serving as core building blocks for international AI start-ups. This marks China's shift from a technology follower to a major builder of the global AI ecosystem, while offering an open and collaborative approach to ensure AI safety and better governance.