Chinese AI lab whose open-weight reasoning models drew global attention. The chat app is capable and free; developers can run the models anywhere.
最适合: Strong open-weight reasoning models at low/no cost.
| 模型 | DeepSeek model family (open weights) |
|---|---|
| 免费计划 | Yes |
| 付费方案 | API is paid; chat largely free |
| API | Yes |
| 自定义指令 | No persistent instruction box. Paste per conversation |
| 记忆 | No cross-chat memory on the standard app |
| 文件上传 | Yes |
| 网络搜索 | Yes |
| 上下文窗口 | Model-dependent |
| 应用 | Web · iOS · Android |
事实最后审核时间 2026-08-10, 功能与方案频繁变动，使用前请以官方网站信息为准。
模型、限制与定价
| 当前模型 | DeepSeek-V4-Pro, flagship reasoning model (deepseek-v4-pro) DeepSeek-V4-Flash, fast, high-concurrency (deepseek-v4-flash, 0731 build) Both released as open weights under the MIT license |
|---|---|
| 上下文窗口 | 1M tokens; 384k max output (both V4 models) |
| 参数 | V4-Pro 1.6T total / 49B active; V4-Flash 284B total / 13B active |
| 输入模态 | Text in and out; thinking mode; tool calls and JSON output |
| 消费者定价 |
|
| API 定价 | deepseek-v4-pro: $0.435 / 1M input (cache miss), $0.87 / 1M output |
定价与模型阵容已核实 2026-08-11 来自 来源 1 ↗ · 来源 2 ↗ · 来源 3 ↗. 此类内容变动频繁，请以供应商官方页面为准。
值得关注
- Open weights, self-hostable
- Visible chain-of-thought reasoning mode