DeepSeek R1

DeepSeek R1, released in January 2025, is an open-weight reasoning model that sparked what became known as the "DeepSeek moment" in AI-demonstrating that frontier-level capabilities could be achieved without the massive training budgets typical of Western AI labs. The model achieves 97.3% on MATH-500 (surpassing OpenAI's o1) and 90.8% on MMLU, showcasing exceptional mathematical reasoning at a fraction of typical development costs.

DeepSeek R1 has won gold medals at both IMO 2025 (International Mathematical Olympiad) and IOI 2025 (International Olympiad in Informatics), demonstrating superhuman performance on elite-level mathematics and competitive programming problems. These achievements highlighted that open-weight models could compete with or exceed proprietary reasoning models on specific benchmarks.

The model's release was significant not just for its technical achievements but for its implications about AI development economics. By achieving ChatGPT-level reasoning capabilities with dramatically lower training costs, DeepSeek R1 challenged assumptions about the resource requirements for frontier AI and accelerated the open-weight movement, inspiring numerous projects to pursue cost-efficient training approaches.