Deepseek R1 (Jan '25)

The R1 inference model released by DeepSeek in January 2025 is specifically optimized for complex reasoning tasks. It adopts an advanced inference architecture and performs excellently in mathematical reasoning, logical analysis, and problem-solving. Although the inference speed is relatively slow (about 25 tokens per second), it demonstrates excellent reasoning ability in pure mathematical benchmark tests. It is suitable for professional applications that require high-quality reasoning and in-depth analysis.