DeepSeek
DeepSeek is an artificial intelligence company founded in 2021 by AI experts from leading tech companies including Google, Microsoft, and OpenAI. The company is dedicated to developing safe, efficient, and cost-effective large language models with the goal of making AI technology more accessible and practical.
Company Development:
- 2021: Company founded, began LLM research and development
- 2023: Launched first commercial API service and DeepSeek-V1 model
- 2024: Released DeepSeek-V2 and Coder-V2 series, gaining wide recognition
- Present: Continuing model optimization and expanding application scenarios
The company offers three main model series:
DeepSeek-Coder-V2:
- High-performance model optimized for code development
- Excellent performance in code generation and understanding
- Supports multiple programming languages and frameworks
- Highest quality score among all DeepSeek models
DeepSeek-V2:
- General-purpose large language model
- Excels in reasoning, knowledge, and general tasks
- Tied for highest quality model alongside DeepSeek-Coder-V2
- Outstanding performance in scientific reasoning and quantitative analysis
DeepSeek-V2.5:
- Latest generation general-purpose language model
- Slightly improved output speed (17 tokens/second)
- Maintains similar performance levels to previous generation
- Optimized for practical application scenarios
Common features across all models:
- Support for 128k large context window
- Full Function Calling support
- JSON mode output support
- First token response time between 1.12-1.15 seconds
- Unified pricing: $0.17 per million tokens (blended price)
Innovative features:
- Context Caching on Disk technology
- Up to 90% API cost savings on cache hits
- Prompt cache pricing: $0.014/million tokens
Performance evaluation:
- Excellent results in MMLU (reasoning and knowledge) tests
- High scores in GPQA (scientific reasoning and knowledge) evaluation
- Outstanding capabilities in mathematical reasoning (MATH) tests
- Excellent performance in code evaluation (HumanEval)
- Strong showing in LMSys Chatbot Arena ELO ratings
Technical Innovation and Research:
- Multiple core technology patents
- Research collaborations with top universities
- Regular publication of technical research papers
- Active participation in open-source community building
Corporate Vision: DeepSeek is committed to providing superior and economical AI solutions to global developers and enterprises through continuous technological innovation and model optimization. The company particularly emphasizes model practicality, reliability, and cost-effectiveness, striving to lower the barriers to AI application.
By striking a balance between model quality, performance, and pricing, DeepSeek is establishing a unique competitive advantage in the AI field while continuously promoting the democratization of AI technology.
DeepSeek-V2.5 is an upgraded version that combines DeepSeek-V2-Chat and DeepSeek-Coder-V2-Instruct. The new model integrates the general and coding abilities of the two previous version ...
1. Introduction We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-eff ...
DeepSeek-V3 is the latest model from the DeepSeek team, building upon the instruction following and coding abilities of the previous versions. Pre-trained on nearly 15 trillion tokens, the reported ...
DeepSeek-R1 is here! ⚡ Performance on par with OpenAI-o1 📖 Fully open-source model & technical report 🏆 MIT licensed: Distill & commercialize freely! ...