FT said Bytedance is Aiming for A Massive AI Model That Might Rival Mythos’ Magnitude

According to persons with knowledge of the situation, the Financial Times claimed on Friday that ByteDance is training an AI model with up to 10 trillion parameters, which may be comparable to the scale of Anthropic’s state-of-the-art Mythos system.

The model’s 10 trillion parameters would make it more than three times larger than the 2.8 trillion-parameter Kimi K3 from the Chinese startup Moonshot AI.

The numerical settings that an AI model learns from data to identify patterns, provide responses, and complete tasks are referred to as its parameters. Although they don’t always indicate capabilities, they are frequently employed as an approximate gage of scale.

Reuters was unable to confirm the report right away. A request for comment from Reuters was not immediately answered by ByteDance.

With 1.6 trillion total parameters, Meituan’s LongCat-2.0 and DeepSeek’s V4-Pro dominated China’s AI market prior to the release of Kimi K3. A number of other domestic competitors have surpassed the trillion-parameter mark.

However, it is challenging to make direct comparisons with top U.S. AI models since businesses like Anthropic and OpenAI do not reveal the number of parameters for systems like Fable, Mythos, or GPT-5.5.

According to industry estimates, Anthropic’s most sophisticated Mythos 5 has over 8 trillion parameters, while Fable 5 has roughly 5 trillion. This means that ByteDance’s latest AI model is nearly as large as Mythos.

The news comes as Chinese tech companies are competing with U.S. rivals to develop more potent systems without making them unaffordable to operate, and they are continuing to speed up their model release cycles to keep up with the growing global AI competition.

According to the FT story, the ByteDance model is presently undergoing pre-training, which usually takes three to six months before it can be refined and made public.