ByteDance has begun early-stage training on an AI model that could rival the scale of Anthropic's most advanced systems, according to three people familiar with the effort. The Chinese tech giant is pre-training a model with as many as 10 trillion parameters—three times the size of Moonshot's Kimi K3, currently the largest Chinese model publicly released.

The parameter count would put it ahead of industry estimates for Anthropic's Mythos 5 (around 8 trillion) and Fable 5 (around 5 trillion), though Anthropic does not officially disclose sizes. Pre-training typically runs three to six months, followed by fine-tuning and a possible release. The final model size is not yet fixed, one source said.

This move underscores how Chinese labs are closing the gap with top US outfits. Recent benchmark results from Moonshot and Alibaba trail only Anthropic's Fable 5 in certain areas. Mythos 5 remains restricted to approved organizations after a temporary security-related ban in June.

Industry insiders say several Chinese labs are training models at Fable 5's scale, but ByteDance is currently the most aggressive in targeting the largest size. Actual capability depends on more than parameter count—data quality and training methods also matter.

Source: Ars Technica