Tensions rose in Silicon Valley this week after DeepSeek, a Chinese AI firm, launched its R1 model, which outperformed top US AI companies, including OpenAI, Meta, and Anthropic, in third-party benchmarks. The anxiety began with DeepSeek v3, which left Meta’s Llama 4 trailing in benchmarks. On Jan. 20, DeepSeek-R1 was released, significantly enhancing model inference capabilities despite limited labeled data. The model performs on par with OpenAI’s GPT-4 in tasks such as mathematics, coding, and natural language reasoning. DeepSeek-R1’s training budget has also drawn attention. According to its published API pricing, it costs RMB 1 ($0.14) per million input tokens (cache hit), RMB 4 ($0.55) per million input tokens (cache miss), and RMB 16 ($2.21) per million output tokens. This pricing is approximately one-thirtieth of OpenAI’s operational costs for GPT-4. Meta’s chief AI scientist, Yan LeCun, commented that DeepSeek’s success demonstrates open-source models are surpassing proprietary ones. [Business Insider, TechNode report]

Related