Byte beats the beam: Big language models are self-researched and accept short-term backwardness

On 6 August, the byte beat CEO Lubo at the mid-2026 plenary meeting stated that the large-language model had been closed by the overseas lead model, but that the byte would be self-researched, do basic work, accept short-term backwardness and maintain long-term optimization. Liang Yubo stated that the insistence on self-study was not due to external pressure, but rather to the desire to delay the team ' s satisfaction and to build a technical base, which was essential for achieving AGI in the long term. The head of the Creative Services Platform Tan Tweet stated that only the SOTA model in the ToB area could take the lead in unlocking high-economic value scenarios, and that video-generation model Feedance 2.0 had pushed the technology from “toys” to serious commercial content production, with a much higher number of days than at weekends. (36 kryptons)