deepseek-ai AI Models

DeepSeekMath-Instruct 7B is a mathematically instructed tuning model derived from DeepSeekMath-Base 7B. DeepSeekMath is initialized with DeepSeek-Coder-v1.5 7B and continues pre-training on math-related tokens sourced from Common Crawl, together with natural language and code data for 500B tokens.

This lightweight 1.5B parameter model excels at fast, cost-effective reasoning tasks with a 131K token context window, making it ideal for developers and researchers seeking efficient inference for question-answering, summarization, and logic-based applications on resource-constrained systems.

DeepSeek-R1-Distill-Qwen-32B is a model distilled from DeepSeek-R1 based on Qwen2.5. It outperforms OpenAI-o1-mini across various benchmarks, achieving new state-of-the-art results for dense models.

Deepseek V3 0324 — DeepSeek open-source language model with competitive performance on reasoning and coding benchmarks.

This 14-billion-parameter distilled model excels at reasoning tasks and general-purpose applications with a massive 131K token context window, making it ideal for developers and researchers seeking cost-effective performance on complex problem-solving, long-document analysis, and multi-turn conversations at just $1.60 per million input tokens.

This distilled reasoning model excels at complex problem-solving and multi-step analysis across a 131K token context, making it ideal for developers and enterprises seeking cost-effective ($2.00/1M input tokens) DeepSeek reasoning capabilities without the computational overhead of the full R1 model.

A cutting-edge language model with 164K token context window designed for researchers, developers, and enterprises requiring advanced reasoning, code generation, and long-document analysis across complex technical and creative tasks.

This reasoning-focused language model excels at complex problem-solving, mathematical analysis, and logical tasks across science, engineering, and research domains, making it ideal for professionals and researchers who need deep analytical capabilities with a large context window for processing extensive documentation and multi-step reasoning.

A high-capacity language model with 164K token context window designed for researchers, developers, and enterprises requiring extended document analysis, complex reasoning tasks, and nuanced content generation across technical and creative domains.

A large language model with 164K token context designed for researchers, developers, and enterprises requiring advanced reasoning, long-document analysis, and complex problem-solving across technical, creative, and analytical domains.

A large language model with 164K token context window designed for developers and researchers who need strong reasoning capabilities, multilingual support, and efficient processing of long documents without the constraints of smaller context windows.

A compact 7-billion parameter language model distilled from DeepSeek R1's reasoning capabilities, optimized for developers and researchers who need efficient reasoning and problem-solving on resource-constrained hardware while maintaining a 131K token context window for processing long documents.

A highly capable frontier language model with 164K token context window designed for researchers, developers, and enterprises seeking strong performance across reasoning, coding, mathematics, and multilingual tasks with efficient inference.

A 671-billion parameter mixture-of-experts language model optimized for research, development, and production applications, suitable for researchers, developers, and organizations seeking high-performance reasoning and coding capabilities with efficient inference costs.

Latest hybrid thinking model from Deepseek

DeepSeek-V3-0324 is the leading non-reasoning model, a milestone for open source

A reasoning model trained with reinforcement learning, on par with OpenAI o1