ByteDance debuts 36b-parameter AI model to rival Google, OpenAI

ByteDance has released three versions of its Seed-OSS-36B open-source AI model on Hugging Face, a major platform for AI developers.

The new model is designed for general and reasoning tasks, and supports long-context processing.

ByteDance, the owner of TikTok and Douyin, said Seed-OSS-36B delivers performance similar to, or better than, models of the same size from competitors such as Alibaba Cloud, Google, and OpenAI, based on recent benchmark tests.

The model has 36 billion parameters, which is smaller than some rival Chinese models.

This move highlights ongoing efforts by Chinese firms to compete with US companies in open-source AI development.

🔗 Source: South China Morning Post

🧠 Food for thought

1️⃣ Chinese AI companies are using open-source releases as a competitive weapon against US tech giants

ByteDance’s decision to release Seed-OSS-36B openly reflects a broader strategic shift by Chinese AI companies to compete through transparency rather than proprietary advantages.

This approach appears to be working, as Stanford’s 2025 AI Index Report shows that while the US still leads in producing top AI models, Chinese models are rapidly closing the performance gap2.

The open-source strategy also delivers measurable business value, with IBM research showing that 51% of businesses using open-source AI tools reported positive ROI compared to just 41% using proprietary alternatives3.

ByteDance’s substantial open-source presence supports this strategy, with their GitHub repository featuring projects like their Multimodal AI Agent Stack that has attracted 17.8k stars and 1.6k forks, demonstrating significant developer engagement4.

By making their AI capabilities freely available, Chinese companies can build developer ecosystems and gather feedback while simultaneously pressuring US competitors who rely on closed, premium models.

2️⃣ Efficiency is becoming more important than raw scale in AI model development

ByteDance’s emphasis on Seed-OSS-36B’s competitive performance despite its “modest” 36 billion parameters signals a significant shift in AI development priorities.

The model’s ability to match or surpass competitors with “parameters dozens of times larger” demonstrates that architectural improvements and training techniques are outpacing the benefits of simply scaling up model size1.