3 weeks ago
ByteDance reportedly trains 10-trillion-parameter AI model to rival Mythos
A big technology company in China called ByteDance is building a very large computer 'brain,' which is called an AI model.
These brains are measured by 'parameters,' which are like the many settings the computer learns to recognize patterns and answer questions.
ByteDance's new model could have up to 10 trillion of these settings.
That would make it more than three times bigger than a model made by another Chinese company called Moonshot AI.
A newspaper called the Financial Times reported this, but not everyone has confirmed it yet.
A model called Mythos 5, made by the company Anthropic, is thought to have about 8 trillion settings, so ByteDance would be close to that size.
The model is still being trained, which usually takes three to six months.
ByteDance is the company that owns the popular video app TikTok.
Right now, companies in China and the United States are racing to build bigger and better AI systems.
ByteDance is training an AI model with as many as 10 trillion parameters, the Financial Times reported on Friday, citing people with knowledge of the matter.
At 10 trillion parameters, the model would be more than three times the size of Moonshot AI's Kimi K3, which has 2.8 trillion parameters.
Industry estimates put Anthropic's Mythos 5 at about 8 trillion parameters and Fable 5 at about 5 trillion parameters, though Anthropic does not disclose official counts.
The model is currently undergoing pre-training, a process that typically lasts three to six months before it can be fine-tuned and released.
Reuters could not immediately verify the report, ByteDance did not respond to a request for comment, and the final parameter count could still change.
- Who
- ByteDance, the Chinese technology company that owns TikTok; Anthropic's Mythos 5 and Moonshot AI's Kimi K3 are cited as comparisons.
- What
- Training of an AI model with up to 10 trillion parameters, a scale close to Anthropic's most advanced Mythos system.
- Where
- China, where ByteDance is based, in the midst of the country's accelerating AI industry race.
- When
- Reported by the Financial Times on Friday; the model is currently undergoing pre-training.
- Why
- Chinese tech firms are speeding up model releases to keep pace with U.S. rivals in the intensifying global AI race.
Scale-focused view
Capability-focused view
Parameter count as a measure of AI strength
Scale-focused view
A higher parameter count signals a more ambitious and powerful model; ByteDance's 10 trillion parameters would approach Anthropic's cutting-edge Mythos system.
Capability-focused view
Parameters are only a rough proxy for scale, not necessarily capability, and because Anthropic and OpenAI do not disclose parameter counts, direct comparisons are unreliable.
Use of AI distillation in model building
Scale-focused view
Chinese labs are racing ahead legitimately by accelerating model releases, as ByteDance positions itself near Mythos-scale with a trillion-parameter model.
Capability-focused view
A senior U.S. official alleged Moonshot AI distilled Anthropic's Fable model while building Kimi K3, and ByteDance founder Zhang Yiming reportedly warned employees against chasing short-term gains through distillation.
Key facts
- Company
- ByteDance
- Reported model size
- Up to 10 trillion parameters
- Status
- Pre-training (typically lasts 3-6 months)
- Reported by
- Financial Times
- Comparison to Kimi K3
- More than 3x larger (2.8 trillion parameters)
- Estimated Mythos 5 size
- About 8 trillion parameters (industry estimate)
- Estimated Fable 5 size
- About 5 trillion parameters (industry estimate)
- Verification
- Reuters could not immediately verify the report









