Buy DeepSeek API access, paid in rial
Most people pick DeepSeek for its price. Its models come close to dearer ones on many tasks, at a fraction of their input price. The table below shows the cost of one concrete job on DeepSeek and three other makers.
One job, several models: cost side by side
Sample job: 10M input tokens and 2M output tokens (say, summarising a few thousand documents). Prices are live; the Intelligence score is there to weigh quality.
| Model | 10M in + 2M out | Intelligence Index |
|---|---|---|
| DeepSeek V4 Flash 0731 | $1.44 | 34.3 |
| DeepSeek V4.1 Flash | $1.97 | 39.5 |
| DeepSeek V4 Pro 0813 | $10.56 | 36.0 |
| Gemini 3.8 Flash | $15.00 | 40.9 |
| GPT-6.1 Sol | $40.00 | 51.8 |
| Claude Sonnet 5.5 | $40.00 | 56.0 |
The same job, as a chart
- DeepSeek V4 Flash 0731$1.44
- DeepSeek V4.1 Flash$1.97
- DeepSeek V4 Pro 0813$10.56
- Gemini 3.8 Flash$15.00
- GPT-6.1 Sol$40.00
- Claude Sonnet 5.5$40.00
Which DeepSeek?
The newest Flash, with image understanding and a compressed cache; for coding agents and long tasks.
- Price
- $0.044/M · $0.765/M
- Context
- 1.05M
- Intelligence Index
- 39.5
- Thinking (reasoning)
- Can be switched on/off
DeepSeek's large model for complex reasoning and big prompts.
- Price
- $0.66/M · $1.98/M
- Context
- 1.05M
- Intelligence Index
- 36.0
- Thinking (reasoning)
- Can be switched on/off
The cheapest input; for bulk text processing and long prompts.
- Price
- $0.016/M · $0.64/M
- Context
- 1.05M
- Intelligence Index
- 34.3
- Thinking (reasoning)
- Can be switched on/off
An experimental version with vision, for documents and charts.
- Price
- $0.216/M · $0.647/M
- Context
- 1.05M
- Intelligence Index
- 34.8
- Thinking (reasoning)
- Can be switched on/off
Open weights and thinking modes
Most DeepSeek models are released with open weights. For you that means two things: the model does not vanish overnight, and if one day you want to run it on your own server, the code you write against the API today stays the same.
The V4 models offer low, high and max thinking. Pick low for short answers to keep output tokens and cost down; DeepSeek's output price is higher than its input price, so controlling thinking length directly moves the bill.
Start in three steps
- 1Sign up on Hormouz with your mobile number.
- 2Buy credit with a bank card and create an API key.
- 3In your OpenAI-compatible code, set the model to deepseek/deepseek-v4.1-flash.
FAQ
- Is DeepSeek really cheaper than GPT and Claude?
- For most tasks, yes; the table above shows the cost of the same job at live prices. Quality differs too, so test both on your own task.
- Is DeepSeek good for Persian?
- Acceptable for most text tasks. For sensitive text, compare its output with another model.
- Do I need a DeepSeek account?
- No. A Hormouz key covers every DeepSeek model, paid with an Iranian bank card.
- Can I switch to another model later?
- Yes. With the same key you change only the model name; your code stays the same.
Prices updated: 2026-10-13