Model

deepseek-r1

9 entries · January 2025 – September 2025

DeepSeek-R1 was an open-weight reasoning model released in January 2025 by the Chinese lab DeepSeek, reported to be comparable to OpenAI's o1 on maths, coding and reasoning benchmarks but published under an MIT licence with a paper describing its training method. It mattered as much for economics as capability: DeepSeek said the underlying V3 model's final training run cost about $5.6 million, far below the budgets assumed for frontier systems, and the release helped trigger a sharp sell-off in US technology stocks — including a record one-day loss for Nvidia — that punctured the assumption that frontier AI required enormous, defensible capital.

Tracks

  • Benchmarks & progress 4
  • Open weights & ecosystem 3
  • Security & misuse 2
  • Models & capabilities 2
  • Money & business 2
  • Safety & alignment 1
  • Ideas & essays 1
  • Government & policy 1
  • Culture & impact 1
  • Compute & infrastructure 1