DeepSeek API Pricing: Cost per Million Tokens, Cache Hit or Miss, and a 1,000-Request Bill
deepseek-flash costs $0.30 per million input tokens on a cache miss and $1.20 per million output tokens at peak hours. Off-peak, both halve. deepseek-v4-pro costs $1.32 and $3.96 at peak. Every rate in this note is from the official DeepSeek pricing page, as of November 6, 2026, in US dollars per million tokens.
The price table
The page lists two models: deepseek-flash (served by DeepSeek-V4.1-Flash-1017) and deepseek-v4-pro (DeepSeek-V4-Pro-0813). Both have a 1M context length and a 384K maximum output.
| Charge, per 1M tokens | Flash off-peak | Flash peak | Pro off-peak | Pro peak |
|---|---|---|---|---|
| Input, cache hit | $0.003 | $0.006 | $0.022 | $0.044 |
| Input, cache miss | $0.15 | $0.30 | $0.66 | $1.32 |
| Output | $0.60 | $1.20 | $1.98 | $3.96 |
Source: DeepSeek API Docs, Models & Pricing, as of November 6, 2026.
Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday, excluding Chinese public holidays. Everything else is off-peak, weekends included. The page states that off-peak rates are half of peak rates.
A cache hit is priced far below a miss: on Flash at peak, $0.006 against $0.30, a fiftieth. The page does not explain what counts as a hit, so treat the cheap row as something to measure on your own traffic rather than assume.
A sample bill
Assume 1,000 requests. Each has 3,000 input tokens, of which 2,000 are a prefix repeated from request to request (a cache hit) and 1,000 are new (a miss), and an answer of 500 tokens. Totals: 2M hit tokens, 1M miss tokens, 0.5M output tokens. These are assumptions for the arithmetic, not measurements.
Flash at peak, line by line:
- Cache hit: 2M x $0.006 = $0.012
- Cache miss: 1M x $0.30 = $0.30
- Output: 0.5M x $1.20 = $0.60
- Total: $0.912
The same 1,000 requests with no cache hits at all (3M miss tokens) cost $0.90 for input plus $0.60 for output, $1.50, so the repeated prefix saves about 39%.
| 1,000 requests | No cache hits | With the 2,000-token prefix cached |
|---|---|---|
| Flash, off-peak | $0.75 | $0.456 |
| Flash, peak | $1.50 | $0.912 |
| Pro, off-peak | $2.97 | $1.694 |
| Pro, peak | $5.94 | $3.388 |
Source: rates from the DeepSeek pricing page, as of November 6, 2026; token counts are the assumptions above.
What moves the bill
Output is the biggest line. On Flash at peak it is $0.60 of the $0.912, from a seventh of the tokens. The page says both models support thinking and non-thinking modes, with thinking as the default, and it does not say how thinking tokens are counted. Look at the output token count of a real response before you trust the 500-token assumption.
Timing is the next lever, a clean factor of two. Model choice comes after: Pro's cache-miss input is 4.4 times Flash's, and its output is 3.3 times.
For scale, the calculator on the home page still shows the April 2023 rate for gpt-3.5-turbo, $0.002 per 1K tokens, or $2 per million. The 3.5M tokens in this sample would be $7.00 at that single rate. That compares prices only; the models are different products. Paste a real prompt into the calculator to get a token count, then swap the counts into the table above.
The page also warns that prices may change, so check it before a budget goes into a spreadsheet.