What happened
Bytedance is training an AI model with up to ten trillion parameters, according to the Financial Times. That size is three times larger than Moonshot's Kimi K3, the current largest Chinese model, and places Bytedance in the same range as Anthropic's Mythos 5, estimated at around eight trillion parameters.
Why it matters
The model is still in pretraining, a phase that typically takes three to six months. Bytedance founder Zhang Yiming has told the company's 2,000-person Seed team to aim for world-leading model capabilities over the long term, signaling the company's ambition to compete at the frontier of AI development.
What to watch
One insider noted that Bytedance has avoided distillation—training on outputs from other companies' models—for over a year, suggesting the company is pursuing an independent path. xAI is also training Grok variants with six and ten trillion parameters on its Colossus 2 cluster, according to Elon Musk.
Summaries like this, in your inbox every morning.
Bytedance's ten-trillion-parameter model represents a significant escalation in China's AI development. The scale places it directly in competition with frontier systems from Western companies like Anthropic, narrowing what was once a clear gap between Chinese and Western AI capabilities. The model's current pretraining phase suggests Bytedance is still months away from releasing or fully evaluating it, meaning the competitive landscape could shift substantially once it emerges.
A notable aspect of Bytedance's approach is its avoidance of distillation for over a year. This suggests the company is building its model from foundational training rather than copying or refining outputs from competitors, which could indicate both confidence in its data and training infrastructure and a strategic choice to develop proprietary techniques. The involvement of founder Zhang Yiming in setting the company's AI ambitions underscores Bytedance's commitment to achieving world-leading capabilities, not merely matching existing systems.
For example, today's edition would include:
AI-summarized, only the topics you pick: one digest a day via Email, LINE, or Slack.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Momentic Inc. launched Mo, an AI agent that opens an app from a URL and prompt, spins up a swarm of agents to…
Anthropic launched Claude Sonnet 5.5, a mid-tier model for everyday tasks, priced at $2 per million input toke…
Oracle group VP Johnnie Konstantas told theCUBE that agentic AI lets agents and sub-agents act as a proxy for…
Qiagen has manually curated biomedical data for more than 25 years with over 150 MD- and PhD-level experts, Bh…
NEAR, the token of the NEAR Protocol, has more than doubled in value over the past two weeks, helped by surgin…
Michael Burry wrote in a Substack chat that Trump's team knows the AI buildout is 'the only thing keeping this…
