The rate card
DeepSeek's API exposes two models: V4.1 Flash (served by the deepseek-flash identifier; the older V4 Flash names now route to it) and V4 Pro (deepseek-v4-pro), which continues with unchanged billing. All figures are USD per 1M tokens, verified against DeepSeek's official API pricing page on October 28, 2026.
| Model | Tier | Input (cache miss) / 1M | Input (cache hit) / 1M | Output / 1M |
|---|---|---|---|---|
| V4.1 Flash1M context, native image input | Off-peak | $0.15 | $0.003 | $0.60 |
| Peak | $0.30 | $0.006 | $1.20 | |
| V4 ProFrontier tier for hard reasoning | Off-peak | $0.66 | $0.022 | $1.98 |
| Peak | $1.32 | $0.044 | $3.96 |
Peak hours: the rate card's second axis
Off-peak rates are exactly half of peak rates. Peak applies Monday to Friday, 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays; every other hour — including all weekend — is off-peak. A workload spread evenly across the week pays roughly 21% more than the off-peak rate; one that avoids both windows pays nothing extra. Schedule non-latency-critical batch work outside the windows and every token halves.
Automatic caching: the deepest discount in the industry
DeepSeek caches prompt prefixes on disk automatically — no SDK changes, no headers, no opt-in. Repeated input tokens bill at the cache-hit rate: $0.003/1M on Flash off-peak, a 50× drop from the $0.15 cache-miss rate. With a stable system prompt and 90% of input as cache hits, your effective Flash input price drops from $0.15 to about $0.018/1M — an ~88% saving on input with no code changes. Keep the reusable part of your prompt at the front and the variable part at the end.
What the bill looks like
Input: 10 × (0.2 × $0.15 + 0.8 × $0.003) = $0.32/day · Output: 5 × $0.60 = $3.00/day
Total ≈ $3.32/day — about $100/month for 450M tokens
The same token volume on a US frontier API runs into the hundreds per month. That gap is the entire DeepSeek pitch — see the head-to-head math in DeepSeek V4.1 Flash vs GPT-6.1 Sol.
Operational notes
- Flat long context. V4.1 Flash charges the same rates across its full 1M-token window — no long-context surcharge, unlike OpenAI's and xAI's banded cards.
- Concurrency limits. DeepSeek documents concurrency ceilings rather than RPM/TPM: its pricing page states 2,500 concurrent requests for Flash and 500 for Pro, with no account tiers.
- OpenAI-compatible endpoint with prepaid billing — no monthly minimum, no per-seat fee.
Default to V4.1 Flash; escalate to V4 Pro only for calls that measurably fail on Flash. Budget either with our bill estimator — and remember to model peak hours.
Quick answers
How much does the DeepSeek API cost?
Verified October 28, 2026: V4.1 Flash is $0.15/$0.003/$0.60 per million tokens (input cache-miss / cache-hit / output) off-peak, doubling at peak. V4 Pro is $0.66/$0.022/$1.98 off-peak, also doubling at peak. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays.
What are DeepSeek's peak hours?
01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays. During these windows every rate doubles — input, output, and cache hits alike. All other hours, including weekends and Chinese public holidays, bill at the off-peak rates, which is why scheduling bulk work into off-peak windows halves DeepSeek bills.
Does DeepSeek have prompt caching?
Yes, and it is automatic: DeepSeek caches prompt prefixes on disk with no SDK changes, no opt-in, and no write fee. Cache hits cost $0.003/1M on V4.1 Flash off-peak — a 98% discount versus the $0.15 cache-miss rate — and the cache is shared across requests with the same prefix.
Is DeepSeek V4 Pro still available?
Yes. DeepSeek initially planned to route V4 Pro traffic to V4.1 Flash, then reversed the plan before it took effect — V4 Pro (API id deepseek-v4-pro) continues with unchanged billing at $0.66/$1.98 off-peak. The legacy deepseek-chat and deepseek-reasoner aliases were retired in July 2026; current ids are deepseek-flash and deepseek-v4-pro.