Pricing hub
DeepSeek Pricing
DeepSeek's public pricing page is especially useful because it exposes cache-hit, cache-miss, and output pricing, which makes repeat workloads easier to model.
Live pricing table
Use this table to compare the published input, cached input, and output rates before you run the calculator.
Input 1,000 / Output 250
| Model | Input / 1M | Cached input / 1M | Output / 1M | Example request | Source |
|---|---|---|---|---|---|
DeepSeek V4 Flash DeepSeek's lower-cost general model with cache-hit discounts for repeated contexts. | $0.140 | $0.0028 | $0.280 | $0.0002 | DeepSeek pricing |
DeepSeek V4 Pro DeepSeek's stronger model with a temporary discount on the public pricing page. The public pricing page currently shows a 75% discount window for V4 Pro through 2026-05-31 15:59 UTC. | $0.435 | $0.0036 | $0.870 | $0.0007 | DeepSeek pricing |
DeepSeek-R1 DeepSeek's reasoning model exposed through the deepseek-reasoner API alias. | $0.550 | $0.140 | $2.19 | $0.0011 | DeepSeek pricing |
How to use this pricing page
- 1. Start with real prompts. Estimate the input and output token counts from the workflow you actually run.
- 2. Check the example cost. The table shows a request-level estimate so you can compare the published rate card with your usage pattern.
- 3. Validate the savings. Open the LLM cost calculator and test a few prompt sizes before you make a budget call.
- 4. Compare the model choice. If you need a second opinion, use one of the linked comparison pages below.
Highlights
- • The cache-hit rate can materially lower repeated-context workloads.
- • V4 Pro currently ships with a public discount window on the pricing page.
- • Useful when you want to compare low-cost reasoning against premium models.
Source: DeepSeek pricing
FAQ
What is the cheapest DeepSeek option here?
DeepSeek V4 Flash is the lower-cost option in the current public pricing table and is usually the first model to compare.
Does DeepSeek show cache pricing?
Yes. The public docs list both cache-hit and cache-miss input pricing, which is important for repeated prompts and agent loops.
Why is V4 Pro shown with a note?
The public docs currently show a 75% discount window for V4 Pro, so the page preserves that context instead of flattening it away.