South China Morning Post reported on Nov. 4, 2026, that DeepSeek engineer Gao Xi defended open-source AI at a World Internet Conference panel in Wuzhen on Tuesday. Gao argued that shared innovation and technical breakthroughs could accelerate artificial general intelligence (AGI). The report did not give the panel’s calendar date.
Gao framed progress as a race against time, rather than a contest between rival companies. The account placed his argument amid US concerns about AI safety and scrutiny in China over potential national-security risks.
Earlier, NeoTeo covered Liu Shengyu’s warning about advanced AI becoming concentrated at Anthropic. Liu’s concern centered on who might control powerful AI; Gao’s reported argument focused on whether shared development could speed progress.
What DeepSeek lists in its R1 release
DeepSeek’s R1 model card says the R1 code repository and model weights are licensed under MIT. DeepSeek’s R1 repository lists DeepSeek-R1, DeepSeek-R1-Zero and six distilled checkpoints as open-sourced. The checkpoints have 1.5B, 7B, 8B, 14B, 32B and 70B parameters.
DeepSeek identifies Qwen2.5 and Llama 3 family models as bases for the distilled versions and lists separate license terms for those underlying models. The MIT statement for R1 code and weights should therefore be read alongside the license terms DeepSeek gives for the distilled models.
The release description names code, weights and model checkpoints. Training data and the full training pipeline are separate parts of model development, beyond the specific artifacts named in that release description.
