What affects cost
- Cache-miss input volume
- Explicit cache-hit assumptions
- Output length
- Thinking or non-thinking mode
- Context length and maximum output
LLM API
DeepSeek API pricing
Useful for cost-sensitive teams evaluating DeepSeek with explicit cache-miss defaults and separately tracked cache-hit pricing.
Decision summary: Use DeepSeek in a shortlist when cache-miss pricing, context length, and output limits fit your workload; only model cache-hit savings with explicit assumptions.
Pricing overview
DeepSeek pricing separates cache-hit input, cache-miss input, and output. StackLens uses cache-miss input by default.
Source-tracked model data
DeepSeek API model prices in one table
Example cost uses 1,000,000 input tokens and 300,000 output tokens at each model's default tracked rate, with no cache discount.
| Model | Input price | Cached input | Output price | Context | Example cost |
|---|---|---|---|---|---|
| DeepSeek V4 Flash non-thinking Stable | $0.14 / 1M | $0.0028 / 1M | $0.28 / 1M | 1M tokens | $0.224 |
| DeepSeek V4 Pro Stable | $0.435 / 1M | $0.0036 / 1M | $0.87 / 1M | 1M tokens | $0.696 |
Cost boundary: this example compares token charges only. It does not assume equal output quality, latency, retry rates, or output length across models.
Lower-cost options from the same provider
Alternative providers or products
Best for
- cost-sensitive experiments
- teams testing provider routing
- large-context workflows after quality testing
Not ideal for
- buyers assuming cache-hit pricing without evidence
- teams that cannot monitor output length
- production routing without quality and latency checks
FAQ
DeepSeek API pricing questions
Does StackLens use DeepSeek cache-hit pricing by default?
No. StackLens uses cache-miss input pricing by default and only applies cache-hit pricing when the user explicitly supplies a cache-hit percentage.
What are the tracked DeepSeek V4 Pro limits?
StackLens stores a 1,000,000 token context length and 384,000 maximum output tokens from the supplied official values.