ByteDance trains massive AI model to take on Anthropic
▼ Summary
– ByteDance is training a large AI model with up to 10 trillion parameters, potentially approaching the size of Anthropic’s most advanced Mythos system.
– The model is at an early pre-training stage, which typically takes three to six months, with final size determined later.
– Anthropic’s Mythos 5 is estimated at about 8 trillion parameters and Fable 5 at 5 trillion, though capability also depends on data quality and training methods.
– Chinese labs like Moonshot and Alibaba are showing strong benchmark performance, trailing only Anthropic’s Fable 5 in some areas.
– Multiple Chinese labs are training models the size of Fable 5, with ByteDance being the most ambitious in pursuing the largest model.
ByteDance is quietly pushing the boundaries of artificial intelligence with a massive new model that could rival the scale of Anthropic’s most advanced systems, signaling just how quickly Chinese labs are closing the distance on their American counterparts.
The Chinese internet giant has begun early-stage training on a model featuring up to 10 trillion parameters, a figure that would make it roughly three times larger than Moonshot’s Kimi K3, currently the biggest Chinese model ever released. That detail comes from three individuals familiar with the project.
Right now, the model is in its pre-training phase, a period that generally stretches from three to six months, according to one of those sources. After that, it would move into fine-tuning and, assuming everything proceeds without major setbacks, eventually see a public release. The final architecture and size, however, won’t be locked in until later.
Anthropic has never published the dimensions of its own models, but industry analysts estimate that its top-tier Mythos 5 carries roughly 8 trillion parameters, while the Fable 5 sits near 5 trillion. Parameter count matters because it defines the raw storage and memory capacity of a model, yet real-world performance hinges on other variables too, including the quality of training data and the methods used to refine the system.
ByteDance’s willingness to chase one of the largest models on the planet underscores a broader ambition among Chinese labs. They are no longer content to simply match the U. S. leaders; they are aiming to surpass them at the highest level of AI development.
Just within the last few weeks, models from Moonshot and Alibaba have posted benchmark scores that trail only Anthropic’s Fable 5 in several categories. Meanwhile, Mythos 5, the flagship Anthropic offering, remains restricted to vetted organizations after a temporary suspension in June tied to safety concerns.
Those watching the industry closely say that several Chinese laboratories are already working on models comparable in scale to Fable 5. But it is ByteDance that currently stands out as the most aggressive, pushing hardest toward the absolute upper limits of model size.
(Source: Ars Technica)
