ByteDance is training an AI model with as many as 10 trillion parameters, a scale that could be close to Anthropic's cutting-edge Mythos system, Reuters reported on Aug. 7, citing the Financial Times (FT).

The FT reported this, citing sources familiar with the matter.

The report said 10 trillion parameters would be more than three times larger than Chinese startup Moonshot AI's Kimi K3, which has 2.8 trillion parameters. AI model parameters refer to numerical settings the model uses to learn patterns from data, generate answers and perform tasks. They are often used as a rough indicator of scale but do not necessarily mean performance.

Before the release of Kimi K3, Meituan's LongCat-2.0 and DeepSeek V4-Pro had 1.6 trillion parameters, and several Chinese rivals also released models with 1 trillion parameters.

Anthropic and OpenAI do not disclose parameter counts for systems such as Fable, Mythos and GPT-5.5, making direct comparisons with major U.S. models difficult. The FT said, based on industry estimates, that Anthropic's latest model Mythos5 could have about 8 trillion parameters and Fable5 about 5 trillion.

The FT report said the ByteDance model is currently in a pretraining stage that typically takes 3 to 6 months, and is expected to be released after fine-tuning.