《From Deepseek 4.0 to Manus Acquisition Withdrawal: Is the AI Gap Between China and the US Widening?》 1️⃣ Deepseek 4.0 Based on publicly available information, we can see a trend: from being on par with closed-source models early last year to only being able to rival them by April this year, the capability gap between Deepseek models and top-tier closed-source models from the US seems to be widening. User tests show significant differences in performance compared to the latest Claude/GPT models. However, Deepseek 4.0 has made a major breakthrough. While it was trained using NVIDIA hardware, its inference now runs on Huawei servers, reducing reliance on overseas GPUs. 2️⃣ Manus Acquisition Withdrawal Meta’s acquisition of the Chinese team Manus was primarily driven by Manus’s agent product and engineering capabilities. But now, with Openclaw + Hermes + Claude’s agent capabilities, Manus no longer holds a competitive edge. It feels like Meta bought in at the peak. If the funds can be withdrawn, it might actually be a positive for Meta? All in all, it seems like the gap in large model capabilities between China and the US is widening. Among the top engineers around me, most are using CC and Codex, with very few opting for domestic models. What do you all think?