DeepSeek Is Running Three AI Races at Once: Catherine Thorbecke

When DeepSeek released its highly anticipated V4 model last month, it landed not with a bang but a whimper. There were no declarations of a new “Sputnik moment” and global markets shrugged.

The muted reaction made sense on the surface. V4 Pro may be the most capable Chinese model released to date, according to an analysis by the US National Institute for Standards and Technology. But the same research found that the gap between DeepSeek and American frontier AI was widening, not shrinking. The model lagged top Silicon Valley offerings by roughly eight months, across tasks including reasoning and math, ...