A large language model with 685B parameters combining high computational efficiency with superior reasoning and agent performance. Features DeepSeek Sparse Attention, reinforcement learning, and tool-use capabilities.

Advanced DeepSeek model with enhanced reasoning capabilities, optimized for complex tasks and extended context understanding.

DeepSeek chat model with strong instruction following and coding, trained on a massive corpus and competitive with closed leaders.

MoE with 30B active per pass; supports "thinking" and standard modes for high-quality reasoning, math, and code.

Instruct-tuned for high-quality conversation and strong general performance across tasks.

17B active parameter multimodal Mixture-of-Experts (MoE) model with 128 experts, tuned for complex business tasks requiring speed and logic.

Open-weight reasoning model with transparent thought tokens; benchmarked near OpenAI o1 in depth-of-reasoning.

17B active parameter model with 16 experts designed to fit on a single Nvidia H100 GPU while offering an industry leading 10M context window.

Compact, instruct-tuned model that's fast and efficient while scoring well in human evals versus larger closed models.

Frontier-scale instruct model with long context and top-tier dialogue quality for complex tasks.

DeepSeek chat model with strong instruction following and coding, trained on a massive corpus and competitive with closed leaders.