DeepSeek
https://deepseek.comChinese AI research lab known for highly efficient open-source models. Their DeepSeek-V3 model achieves top performance at fraction of training cost.
Models by DeepSeek (3)
DeepSeek-V3
128K ctxHighly efficient 671B MoE model trained on 14.8T tokens. Achieves top benchmark scores at fraction of typical training cost.
DeepSeek-R1
64K ctxReasoning-focused model achieving strong performance on math and coding benchmarks through reinforcement learning.
DeepSeek-V3.2
DeepSeek-V3.2 is the official successor to V3.2-Exp, designed as a reasoning-first model built for agents. It is positioned as a daily driver with performance at the GPT-5 level, balancing inference and length. The V3.2-Speciale variant pushes the boundaries of reasoning capabilities, rivaling Gemini-3.0-Pro, and is currently available only via API. The model excels in complex tasks, achieving gold-level results in prestigious competitions such as IMO, CMO, ICPC World Finals, and IOI 2025.