ByteDance is training an AI model with up to 10 trillion parameters

ByteDance, the parent company of TikTok, is training an AI model with up to 10 trillion parameters, according to a Financial Times report. That would be the largest known model size from a Chinese company – approaching Western frontier models such as Anthropic's Mythos 5 at an estimated ~8 trillion parameters.

The model is being built in ByteDance's Seed division. Pre-training such giants typically takes three to six months; a release is expected no earlier than 2027. Reuters confirmed the reports on 7 August 2026.

The number alone says nothing about quality – parameter count is not the same as intelligence. But it is a clear strategic signal: China is catching up in the pure compute and scaling race. The gap between Western and Chinese frontier models is increasingly measured in months, not generations.

For European companies the development is doubly relevant. First, competition is growing – and with it potentially price pressure on capable models. Second, the governance question sharpens: where does the model come from, where does the data sit, which rules apply? Especially for regulated industries, the provenance of AI systems is becoming a selection criterion.