[{"data":1,"prerenderedAt":30},["ShallowReactive",2],{"nr-en-bytedance-trainiert-10-billionen-parameter-modell":3},{"slug":4,"title":5,"dek":6,"date":7,"time":8,"publishedAt":9,"updated":10,"updatedAt":10,"dateFmt":11,"updatedFmt":10,"kind":12,"tier":13,"author":14,"authorName":15,"topics":16,"tracker":22,"trackerLabel":23,"headlineStat":24,"image":25,"ogImage":26,"imageAlt":5,"csv":10,"minutes":27,"words":28,"html":29},"bytedance-trainiert-10-billionen-parameter-modell","ByteDance Trains 10-Trillion-Parameter Model – China's AI Offensive Against the West","TikTok's parent company is building one of the world's largest language models. The move shows how aggressively Chinese labs are closing the gap with Anthropic and OpenAI.","2026-08-07","18:51","2026-08-07T18:51:00+02:00","","August 7, 2026","news","standard","ideal-syka","Ideal Syka",[17,18,19,20,21],"AI models","China","ByteDance","Anthropic","Language models","\u002Fstand-der-ki","AI progress","10 trillion parameters","\u002Fnewsroom\u002Fimg\u002Fbytedance-trainiert-10-billionen-parameter-modell.webp","\u002Fog-nr\u002Fbytedance-trainiert-10-billionen-parameter-modell.en.png",3,507,"\u003Cp>ByteDance is currently training an AI model with up to \u003Cstrong>10 trillion parameters\u003C\u002Fstrong> – three times larger than the biggest Chinese model released to date and potentially competitive with Anthropic's most advanced architecture. Ars Technica reports this based on three people with knowledge of the matter. The model is still in early pre-training, a phase that typically lasts three to six months.\u003C\u002Fp>\n\u003Ch2>The essentials\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>ByteDance is training a \u003Cstrong>10-trillion-parameter model\u003C\u002Fstrong> – significantly larger than Moonshot's \u003Cstrong>Kimi K3\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>Anthropic's \u003Cstrong>Mythos 5\u003C\u002Fstrong> is estimated at about \u003Cstrong>8 trillion parameters\u003C\u002Fstrong>, \u003Cstrong>Fable 5\u003C\u002Fstrong> at about \u003Cstrong>5 trillion\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>The ByteDance model is being developed independently – \u003Cstrong>without distillation\u003C\u002Fstrong> from competitor models\u003C\u002Fli>\n\u003Cli>Seed, ByteDance's AI team led by former Google DeepMind scientist \u003Cstrong>Wu Yonghui\u003C\u002Fstrong>, has about \u003Cstrong>2,000 employees\u003C\u002Fstrong> worldwide\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>China's catch-up accelerates\u003C\u002Fh2>\n\u003Cp>The move underscores how quickly Chinese labs are narrowing the gap with US providers. In recent weeks alone, models from Moonshot and Alibaba have shown strong benchmark performance – in some areas only slightly behind Anthropic's Fable 5. While Mythos 5 is available only to approved organizations following a security ban in June, multiple Chinese labs are already working on Fable 5-sized models. ByteDance positions itself as the most ambitious initiative.\u003C\u002Fp>\n\u003Cp>What's distinctive: ByteDance pursues an \u003Cstrong>independent development approach\u003C\u002Fstrong> without so-called model distillation – that is, without compressing and adopting insights from existing competitor models. This strategy has been in place for over a year. Founder \u003Cstrong>Zhang Yiming\u003C\u002Fstrong> emphasized in an internal meeting two weeks ago with the Seed team that only genuine independent development leads to superior models.\u003C\u002Fp>\n\u003Ch2>ByteDance: From TikTok conglomerate to AI factory\u003C\u002Fh2>\n\u003Cp>ByteDance has kept a low profile on AI announcements – unlike many Chinese competitors. Yet behind the scenes, the company is investing massively: Over three years, ByteDance has invested more in AI than any other Chinese tech giant. The company is expanding data centers, recruiting researchers, and has significantly scaled its cloud business Volcano Engine, which sells AI solutions to enterprises.\u003C\u002Fp>\n\u003Cp>ByteDance's existing AI flagships already demonstrate strength: the video generation model \u003Cstrong>SeeDance\u003C\u002Fstrong> ranks among the world's most advanced, while the consumer model \u003Cstrong>Doubao\u003C\u002Fstrong> has \u003Cstrong>324 million monthly active users\u003C\u002Fstrong> and is described as China's most popular. The Seed team comprises not only core researchers but also infrastructure engineers, data labelers, and translators – a structure oriented toward long-term, independent development.\u003C\u002Fp>\n\u003Ch2>What this means for German companies\u003C\u002Fh2>\n\u003Cp>The ByteDance initiative illustrates a fundamental shift in global AI competition: Chinese providers are no longer merely seeking parity with Western models but aiming for superiority. For German companies relying on Western AI infrastructure, this could mean mid-term pressure on prices and availability – yet also create new diversification options if Chinese models become regulatorily accessible. Important caveat: raw parameter count says nothing about actual performance. What matters is data quality, training methods, and practical applicability. Whether ByteDance's 10-trillion model truly rivals Anthropic will only become clear after release.\u003C\u002Fp>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Farstechnica.com\u002Fai\u002F2026\u002F08\u002Fbytedance-trains-massive-ai-model-in-bid-to-rival-anthropic\u002F\">Ars Technica\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cem>Editorially owned by \u003Ca href=\"\u002Fen\u002Fautor\u002Fideal-syka\">Ideal Syka\u003C\u002Fa>. Sources and method: \u003Ca href=\"\u002Fen\u002Fredaktion\">Newsroom & method\u003C\u002Fa>. Tips and corrections: \u003Ca href=\"mailto:ai@i6eal.de\">ai@i6eal.de\u003C\u002Fa>.\u003C\u002Fem>\u003C\u002Fp>\n",1786121789130]