ByteDance reportedly trains a model of up to 10 trillion parameters

The reports reported, and a report relayed, that ByteDance is pre-training a model with up to 10 trillion parameters, with fine-tuning and release potentially several months away. ByteDance did not comment, and a report said it could not independently verify the report.

Why it matters: If accurate, the project would signal continued pursuit of extreme model scale. Parameter count alone does not establish quality, efficiency, safety or commercial readiness. This remains an unconfirmed report, not a release.