通过 MuAPI 使用 DeepSeek V4.1 Flash，进行多模态推理、图像理解，并灵活调整思考力度。 DeepSeek V4.1 Flash 是a fast multimodal reasoning model with a 1M-token context window, native image understanding, and configurable reasoning effort.
通过 MuAPI 使用 DeepSeek V4.1 Flash，进行多模态推理、图像理解，并灵活调整思考力度。
关于此模型
DeepSeek V4.1 Flash 是a fast multimodal reasoning model with a 1M-token context window, native image understanding, and configurable reasoning effort.
成本分析
| 提供商 | 费用 | 备注 |
|---|---|---|
| muapiapp | Official DeepSeek token rates; peak/off-peak and cache-sensitive | Off-peak: $0.15/M uncached input, $0.003/M cached input, $0.60/M output. Peak rates are 2×. A $0.0001 minimum reservation 是reconciled after usage. |
Off-peak: $0.15/M uncached input, $0.003/M cached input, $0.60/M output. Peak rates are 2×. A $0.0001 minimum reservation 是reconciled after usage.
** 竞品价格根据相似模型架构和使用层级估算。
配置参数
| 参数 | 类型 | 描述 | 默认值 |
|---|---|---|---|
| 提示词 | string | 提供给模型的用户消息或指令。 | Explain how a mixture-of-experts model works. |
| 图像 URL | string | 可选的 public image URL for multimodal analysis. | undefined |
| 系统提示词 | string | 可选的 system-level instruction. | You are a careful software engineering assistant. |
| 推理强度 | 枚举（7 个选项） | Thinking effort. Omit to use the model default; none and minimal disable thinking. | - |
提供给模型的用户消息或指令。
Explain how a mixture-of-experts model works.可选的 public image URL for multimodal analysis.
undefined可选的 system-level instruction.
You are a careful software engineering assistant.Thinking effort. Omit to use the model default; none and minimal disable thinking.
-开发者文档
/api/v1/deepseek-v4-1-flash 发送 POST 请求，或通过 /api/v1/deepseek-v4-1-flash/stream.image_url 以进行视觉理解。reasoning_effort to none, minimal, low, medium, high, xhigh, or max；省略此项则使用模型默认值。常见问答
可以。 您可以provide an optional public image URL alongside your prompt.
可以。 将 `reasoning_effort` to one of the supported levels. `none` and `minimal` disable thinking.
可以。 Live streaming 是available at `/api/v1/deepseek-v4-1-flash/stream`.