In August 2026, the technological landscape of the global artificial intelligence market is undergoing fundamental changes. Chinese giant ByteDance, world-famous as the creator of the social network TikTok, officially confirms its massive ambitions in the field of large language models (LLM). According to authoritative sources, including Financial Times, the company is in the final stages of pre-training its own AI model of colossal scale — with 10 trillion parameters.

Technological race: 10 trillion vs 8 trillion

The development of a model with 10 trillion parameters places ByteDance on par with leading global players, such as the American startup Anthropic. For comparison, Anthropic's flagship model, Mythos 5, is estimated by experts at 8 trillion parameters, while its more compact version, Fable 5, is at 5 trillion. Thus, the Chinese model is not just catching up but is claiming leadership in terms of parameter volume, which theoretically allows it to process and store unprecedented amounts of data in memory.

However, experts emphasize that the number of parameters is not the only and not always the determining factor of success. The quality of input data, model architecture, and training methods play a critical role. Unlike many competitors who resort to distillation (simplification) of others' developments, ByteDance is betting on its own resources and unique developments, striving to create a model that surpasses Western counterparts in generation quality and logical thinking.

Strategy of "Long-term Dominance"

ByteDance founder Zhang Yiming has formulated a clear strategy for his team: short-term lag does not matter in light of long-term goals. A team of 2,000 specialists, including engineers, researchers, data labeling specialists, and translators, is working on the creation of the AI model within the company. Work is being carried out both in China and in international offices, allowing the company to use global talents and data.

Currently, the model is in the pre-training stage, which, according to insiders, takes from three to six months. This will be followed by a fine-tuning stage, necessary for adapting the model to specific tasks and safety. Only after completing all stages will the model be released for commercial use, which will likely happen at the end of 2026 or early 2027.

Contradictory data

There are disagreements in the AI industry regarding how much the parameter count correlates with the actual capabilities of the model. On the one hand, proponents of scaling laws argue that increasing parameters to 10 trillion will inevitably lead to the emergence of new, previously unavailable capabilities (emergent properties). On the other hand, critics point to the example of the Chinese model Kimi 3 from Moonshot, which with slightly less than 3 trillion parameters already demonstrates outstanding results, being considered one of the most efficient in China.

Moreover, exact data on the parameters of Anthropic models (Mythos 5 and Fable 5) are estimated, as the company does not disclose this information officially. This creates a certain uncertainty in comparison: ByteDance may be ahead of competitors in "paper" characteristics, but real superiority will only become clear after public benchmarks and tests.

Global consequences and competition

The entry of ByteDance's model into the market could radically change the balance of power in the AI race. Given the integration with the TikTok ecosystem, the new model will get instant access to a huge amount of user data and use cases, which will accelerate its learning and adaptation. This creates a direct threat to OpenAI and Google, which have dominated the segment of ultra-powerful models so far.

Furthermore, ByteDance's success could become a catalyst for other Chinese technology companies that are also actively investing in AI. It is expected that in the coming months we will see a surge of announcements of new models from China, which will lead to even stiffer competition in the global market.