Timeline
- The US Department of Homeland Security publishes an order: federal civilian agencies must stop procuring DeepSeek products and remove from their IT systems both those products and third-party products built on DeepSeekLaws & regulation · Geopolitics & defense · Cybersecurity · Compute export controls · federalregister.gov ↗
- DeepSeek open-sources its production inference engine DeepInfer under Apache 2.0 and reports that a single cluster running DeepSeek V4.1 Flash serves 155 trillion tokens a day, earning $38 per GPU-hour at the off-peak rateSoftware development · Business & markets · Internet & infrastructure · Open weights · github.com ↗
- ByteDance Seed: with chunked KV-cache compression, open-weight models such as DeepSeek-V4 show long-context retrieval accuracy differing by up to 40 percentage points across “phases” — shifts of the input by a few tokensSoftware development · Science · Open weights · arxiv.org ↗
- DeepSeek-R1: a cheap open reasoning model; on 27 January NVIDIA shares fall ~17%Finance & money · Business & markets · Open weights · api-docs.deepseek.com ↗
- DeepSeek-V3 was trained on 2,048 H800 chips cut down for China: 2.788 million GPU hours, about $5.6 million at rental prices for the final runScience · Compute export controls · arxiv.org ↗