DeepSeek-R1

With supporting stream, DeepSeek-R1 is an open-source large language model (LLM) from the AI startup DeepSeek, designed for advanced reasoning tasks like mathematics, coding, and general knowledge. Trained using reinforcement learning (RL) followed by supervised fine-tuning, it self-evolves and refines its outputs for clarity and accuracy. The model achieves 79.8% on AIME 2024 math tests, a 2,029 Codeforces rating (outperforming 96.3% of programmers), and 90.8% accuracy on MMLU benchmarks, rivaling OpenAI's o1 model. Fully open-source under the MIT license, it is accessible via Hugging Face, encouraging collaboration. Its API service, DeepSeek Reasoner, offers competitive pricing, making it a cost-effective and high-performing solution for developers and researchers alike.

Published on 2025-03-15 by @swift-ai.