Back to models
DeepSeek-V3
Flagship modelNewRecommended
DeepSeek's flagship mixture-of-experts (MoE) model, matching GPT-4o performance with extreme efficiency.
| Context window | 64,000 tokens |
| Max output tokens | 8,192 tokens |
Pricing
| Type | Input | Output |
|---|---|---|
| Standard | $0.28 / 1M tokens | $1.11 / 1M tokens |
Parameters
| Parameter | Default | Range |
|---|---|---|
| Temperature | 1 | 0 to 2 |
| Top P | 1 | 0 to 1 |
| Frequency penalty | 0 | -2 to 2 |
| Presence penalty | 0 | -2 to 2 |
| Max stop sequences | 4 | — |
Supported features
Capabilities
- Streaming
- Function Calling
- Structured Output
- Json Mode
- System Prompt
Endpoints
- Chat Completions
Use cases
- text-chat
- function-calling
Model specifications
| Model ID | deepseek-v3 |
| Family | deepseek-v3 |
| Access type | Closed |
| Input modalities | text |
| Output modalities | text |