API Cost Check
DeepSeek API cost calculator
DeepSeek prices input, output and cache hits separately and caches prompt prefixes automatically: there is nothing to mark, no write fee in our data, and a hit is billed at 2%–3.33% of the input price, one of the deepest cache discounts we track. Because the cache is the main lever, put the fixed part of every prompt first so hits happen on every request.
There is no Batch API in our data, so the Batch column below repeats the standard price with a flag. For the same traffic DeepSeek V4 Pro costs 4× as much as DeepSeek V4 Flash; both share the same context window, so choose by answer quality on your own prompts. DeepSeek has at times offered time-of-day discounts; those are not in our data.
The table prices one workload on each model: 10,000 requests a day with 6,000 input and 500 output tokens per request. Columns: no caching; 70% of input read from the cache and 5% written to it; every request sent through the Batch API. For the same traffic DeepSeek V4 Pro costs 4× as much as DeepSeek V4 Flash.
| Model | Standard 10,000 req/day · 6,000 in · 500 out Calculate with your numbers → | With prompt caching 10,000 req/day · 6,000 in · 500 out · cache 70% Calculate with your numbers → | Batch API 10,000 req/day · 6,000 in · 500 out · batch Calculate with your numbers → |
|---|---|---|---|
| DeepSeek V4 Flash | $720/mo | $350/mo | $720/mo no batch |
| DeepSeek V4 Pro | $2,970/mo | $1,362/mo | $2,970/mo no batch |
Estimates from published per-token rates; check official pricing before budgeting. Data synced from the LiteLLM price list 2026-10-12 08:00 UTC.
Cost per 1,000 requests
For 1,000 requests of 6,000 input and 500 output tokens, DeepSeek V4 Flash costs $2.40 at standard rates, $1.17 with caching and $2.40 through the Batch API; DeepSeek V4 Pro costs $9.90, $4.54 and $9.90. Multiply by your own request count, or put your traffic into the calculator.
How to cut the bill, in order
- Keep the fixed prefix (instructions, examples, documents) identical across requests; a single changed byte at the start misses the cache for the whole prompt.
- Use Flash as the default and escalate to Pro only for requests that fail a quality check.
- Cap output length and avoid asking for reasoning where a short answer is enough: reasoning tokens are billed as output.
- Count prompts with the token counter and check your hit rate in the caching calculator.
- Put your traffic into the calculator; with no batch tier, caching and output length are the only levers.
When does caching pay off for your traffic? Use the prompt caching calculator →
FAQ
Is there a cache write fee on DeepSeek?
Not in our data: a miss is billed as normal input and later hits at the cache-hit price. Hits require an identical prefix, so keep the fixed part byte-for-byte stable.
Why is there no Batch column discount?
DeepSeek lists no batch price, so the column shows the standard price with a no-batch flag. For offline jobs the cache still applies when prompts share a prefix.
Does DeepSeek charge for reasoning tokens?
Yes. Tokens generated while reasoning before the answer are billed as output tokens. Budget for them by measuring real responses or by raising the output tokens per request in the calculator.
Flash or Pro?
Flash for volume, latency and routine tasks; Pro for hard reasoning, code and agents. Both have the same context window, so the only trade is quality per dollar: run your prompts on both before committing.