ByteDance is targeting a mega AI model nearing Anthropic's Mythos, according to FT

Via FT

Frontier-model ambitions from large Chinese platforms have followed a familiar sequence: a capability claim surfaces via press reporting, the market then waits for an actual release or benchmark disclosure before repricing anything, since self-described parity with leading Western models has historically been asserted well before independent evaluation confirms it. The actors here matter: ByteDance has prior form as a fast-follower, converting scale in distribution and capital into compressed development timelines, while the comparison target being an Anthropic-class model sets the bar at the frontier rather than at the open-weights tier where Chinese labs have more often competed. The distinction worth drawing is between training a model and monetising it: claims of this kind have tended to matter first for the domestic peer set and the compute supply chain feeding it, and only later for global competitive positioning. Near-term follow-ons are confirmation from the company itself, any details on parameters or training compute, and whether the claim holds up under third-party benchmarking. As an unconfirmed media report on a private company, the signal is directional and headline-driven rather than fundamental.