DeepSeek
深度求索推出的开源 AI 模型系列。以极高性价比和强推理能力闻名全球，DeepSeek-R2 在数学和编程能力上对标 OpenAI o4。采用 MoE（混合专家）架构，推理效率远超同参数规模模型。完全免费使用，API 价格仅为同类的 1/10。
DeepSeek's open-source AI model series. Globally renowned for extreme cost-efficiency and strong reasoning — DeepSeek-R2 rivals OpenAI o4 in math and coding. Uses MoE (Mixture of Experts) architecture for vastly superior inference efficiency. Completely free to use, with API pricing at 1/10 of competitors.
🧠 可用模型Available Models
| 模型 | Model | 上下文 | Context | 发布 | Release | 特点 | Highlights |
|---|---|---|---|---|---|---|---|
| DeepSeek-R2 HOT | 256K | 2026.02 | 第二代旗舰推理模型，链式思考+自适应推理，数学/编程对标 o4，开源可商用 | ||||
| DeepSeek-V4 NEW | 256K | 2026.05 | 第四代通用模型，MoE架构，综合能力大幅提升，多模态理解 | ||||
| DeepSeek-V3.1 | 128K | 2025.06 | V3 增强版，推理能力提升，编程和长文本表现佳 | ||||
| DeepSeek-R1-Distill | 128K | 2025.01 | 蒸馏推理模型，7B-70B 多种规格，适合本地部署 | ||||
| DeepSeek-R2 HOT | 256K | 2026.02 | 2nd-gen flagship reasoning, chain-of-thought + adaptive inference, rivals o4 in math/coding, open source | ||||
| DeepSeek-V4 NEW | 256K | 2026.05 | 4th-gen general model, MoE architecture, major capability leap, multimodal understanding | ||||
| DeepSeek-V3.1 | 128K | 2025.06 | Enhanced V3, improved reasoning, strong at coding and long text | ||||
| DeepSeek-R1-Distill | 128K | 2025.01 | Distilled reasoning models, 7B-70B variants for local deployment |
💰 价格方案Pricing Plans
Web / App
- 完全免费使用
- 联网搜索
- 文件上传（图片/文档）
- 1M 上下文支持
API — V4
- 输入极低价
- 输出 $0.98/1M
- 适合批量任务
API — R2
- 推理模型 API
- 输出 $2.00/1M
- 缓存命中更低价
🎯 适用场景Use Cases
💻 编程开发Coding
R1 模型在算法竞赛和工程编码中表现顶尖，支持多语言
R1 model excels in competitive programming and engineering coding, multilingual
📊 数据分析Data Analysis
128K上下文上传CSV/Excel，自动分析和可视化
128K context for CSV/Excel uploads, auto analysis and visualization
🎓 学术研究Academic Research
数学证明、论文润色、文献综述，R1 推理能力领先
Math proofs, paper polishing, literature review — R1 leads in reasoning
🏢 企业部署Enterprise Deployment
Apache 2.0 开源许可，可自托管，数据不外传
Apache 2.0 license, self-hostable, data stays in-house
🔗 相关工具Related Tools
❓ FAQ
DeepSeek 真的完全免费吗？
是的，Web 和 App 端完全免费，包括联网搜索和文件上传功能。API 调用按 tokens 计费，但价格仅为 GPT-4o 的 1/10-1/20。
DeepSeek-R2 和 V4 怎么选？
日常对话、内容生成用 V4 即可。遇到复杂数学、编程或逻辑推理问题时切换到 R2（Deep Think 模式），模型会自动进行链式推理。R2 的自适应推理能在简单任务上节省 Token。
Is DeepSeek really completely free?
Yes, Web and App are fully free including web search and file uploads. API is priced per token but at 1/10 to 1/20 the cost of GPT-4o.
R2 vs V4 — which to use?
Use V4 for daily chat and content generation. Switch to R2 (Deep Think mode) for complex math, coding, or logic problems — the model auto-performs chain-of-thought reasoning. R2's adaptive inference saves tokens on simple tasks.