Here is the number that should stop you mid-scroll: 10 trillion. That is how many parameters ByteDance is reportedly targeting for its next major AI model — roughly three times the size of the largest Chinese model currently in public existence, and potentially bigger than anything Anthropic has deployed.
To put that in perspective, industry estimates peg Anthropic's most advanced system, Mythos 5, at around 8 trillion parameters. ByteDance is not just trying to close the gap with American AI labs. It is trying to leap over them.
Now, a quick reality check before the hype spiral begins. The model is still in early pre-training, a phase that typically runs three to six months before fine-tuning even begins. The final parameter count has not been locked in yet. And bigger does not automatically mean better — data quality, training methodology, and architecture choices all have enormous influence on what a model can actually do. Parameters are potential, not performance.
Still, the ambition here is hard to overstate. ByteDance has been quietly pouring more money into AI infrastructure than any other Chinese tech company, building out data centers, hiring researchers globally, and doubling down on its cloud business through its enterprise-facing Volcano Engine unit. Its model development team, called Seed, runs roughly 2,000 people and is led by a former Google DeepMind scientist.
What makes this push particularly interesting is how ByteDance is doing it. Unlike several Chinese competitors, the company has reportedly avoided model distillation — the practice of training a smaller model to mimic the behavior of a larger one — for more than a year. That approach tends to produce faster results, which may explain why some rivals have moved quicker. But it also means ByteDance is building from scratch, with no shortcuts baked in. If this model lands well, it will be harder to dismiss.
ByteDance is not exactly a stranger to AI wins. Its SeeDance video generation model is considered among the best in the world. Its consumer chatbot, Doubao, has 324 million monthly active users in China, making it the most used AI assistant in the country by a wide margin. The company knows how to ship products people actually use.
The broader context matters too. Chinese labs have been closing the benchmark gap with American ones at a pace that would have seemed implausible two years ago. Models from Moonshot and Alibaba now trail only Anthropic's second-tier Fable 5 in some evaluations. Meanwhile, Anthropic's most powerful system, Mythos 5, is restricted to approved organizations after a brief government-mandated pause in June over security concerns.
If ByteDance sticks the landing on a 10 trillion parameter model, it would not just be a flex. It would signal that the AI race has genuinely become a two-continent competition, with no clear leader locked in.
POLICY
Meta ordered to pay nearly 1 billion dollars in child safety case
A single U.S. state just handed Meta a nearly $1 billion bill, and the legal theory behind it is one the company should be worried about far beyond New Mexico's borders.
A Santa Fe district court ruled this week that Meta must pay $567 million into an abatement fund as part of the second phase of New Mexico's landmark child safety case. Combined with $375 million in civil penalties from the first phase, Meta's total tab in this one case alone has climbed to $942 million. The court found that Meta's platforms constitute a "public nuisance" and are a "significant contributing cause" of a teen mental health crisis affecting public health and safety across the state.
The public nuisance framing is worth paying attention to. It is the same legal approach that states used to extract massive settlements from opioid manufacturers, and it is designed to hold companies accountable for broad societal harm rather than individual injuries. If that logic holds up on appeal — and Meta has already said it plans to fight the ruling — it opens a path for other states to pursue similar claims without needing plaintiffs to prove specific, individual damages.
New Mexico Attorney General Raul Torrez framed the ruling as a message to the broader industry. His statement made clear that this was about establishing that profiting from practices that harm young people carries real financial consequences. That is a pointed framing, and other state attorneys general are almost certainly watching.
The $567 million abatement fund is earmarked for practical interventions: awareness campaigns, prevention programs, mental health screening, and treatment for people harmed by the platforms. That is a different outcome than a fine that disappears into general government revenue. The money is supposed to fix something.
Meta's response was predictable and consistent with its previous posture on child safety litigation. A spokesperson said the company disagrees with the ruling, called the characterizations inaccurate, and pointed to its existing safety efforts. The company has repeatedly argued that it works hard to remove bad actors and harmful content, even while acknowledging the difficulty of doing so at scale.
The problem for Meta is that "we tried" is not landing as a defense in court. And with nearly $1 billion on the line from a single state, the financial stakes of that legal strategy are becoming very real.
This ruling lands at a particularly uncomfortable moment for the company. Meta has been leaning into AI features across Instagram, Facebook, and WhatsApp, many of which are aimed squarely at younger users. Every time the company expands its reach with that demographic, it also expands its exposure to exactly this kind of scrutiny.
The appeal will likely drag this out for years. But the precedent New Mexico is building, case by case and phase by phase, is not going away.
Source: The Verge
A Santa Fe district court ruled this week that Meta must pay $567 million into an abatement fund as part of the second phase of New Mexico's landmark child safety case. Combined with $375 million in civil penalties from the first phase, Meta's total tab in this one case alone has climbed to $942 million. The court found that Meta's platforms constitute a "public nuisance" and are a "significant contributing cause" of a teen mental health crisis affecting public health and safety across the state.
The public nuisance framing is worth paying attention to. It is the same legal approach that states used to extract massive settlements from opioid manufacturers, and it is designed to hold companies accountable for broad societal harm rather than individual injuries. If that logic holds up on appeal — and Meta has already said it plans to fight the ruling — it opens a path for other states to pursue similar claims without needing plaintiffs to prove specific, individual damages.
New Mexico Attorney General Raul Torrez framed the ruling as a message to the broader industry. His statement made clear that this was about establishing that profiting from practices that harm young people carries real financial consequences. That is a pointed framing, and other state attorneys general are almost certainly watching.
The $567 million abatement fund is earmarked for practical interventions: awareness campaigns, prevention programs, mental health screening, and treatment for people harmed by the platforms. That is a different outcome than a fine that disappears into general government revenue. The money is supposed to fix something.
Meta's response was predictable and consistent with its previous posture on child safety litigation. A spokesperson said the company disagrees with the ruling, called the characterizations inaccurate, and pointed to its existing safety efforts. The company has repeatedly argued that it works hard to remove bad actors and harmful content, even while acknowledging the difficulty of doing so at scale.
The problem for Meta is that "we tried" is not landing as a defense in court. And with nearly $1 billion on the line from a single state, the financial stakes of that legal strategy are becoming very real.
This ruling lands at a particularly uncomfortable moment for the company. Meta has been leaning into AI features across Instagram, Facebook, and WhatsApp, many of which are aimed squarely at younger users. Every time the company expands its reach with that demographic, it also expands its exposure to exactly this kind of scrutiny.
The appeal will likely drag this out for years. But the precedent New Mexico is building, case by case and phase by phase, is not going away.
Enjoyed this?
Get stories like this delivered every Tuesday — free.