· CostMyTokens · Analysis · 5 min read
DeepSeek API Pricing 2026: Is It Really the Cheapest?
DeepSeek API pricing for deepseek-flash and deepseek-v4-pro from the official price list — peak and off-peak rates, cache-hit pricing, monthly costs, and an honest answer to the cheapest question.
DeepSeek’s reputation rests on one claim: flagship-class models at a fraction of the price. Its official price list is also the most unusual of any major provider — the price depends on when you call and whether your prompt hits the cache. This page lists the official rates for both models DeepSeek currently sells, works out monthly costs at peak and off-peak, and then answers the cheapest question with numbers. For the head-to-head against closed flagships, see our DeepSeek V4 Pro vs closed models comparison.
Disclosure: this article contains no affiliate links. Where CostMyTokens does use affiliate links elsewhere, they never affect order — every table is sorted by price only. Read our disclosure.
DeepSeek API prices (official, per 1M tokens)
DeepSeek’s current price list has exactly two models. Both are priced twice: peak and off-peak. Off-peak is always half of peak.
| Model | Input (cache miss) / 1M | Input (cache hit) / 1M | Output / 1M | Context | Source |
|---|---|---|---|---|---|
| deepseek-flash (DeepSeek-V4.1-Flash-1017) | $0.15 off-peak / $0.30 peak | $0.003 off-peak / $0.006 peak | $0.60 off-peak / $1.20 peak | 1,000,000 | official pricing, verified 2026-10-26 |
| deepseek-v4-pro (DeepSeek-V4-Pro) | $0.66 off-peak / $1.32 peak | $0.022 off-peak / $0.044 peak | $1.98 off-peak / $3.96 peak | 1,000,000 | official pricing, verified 2026-10-26 |
Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays. Everything else is off-peak — including all of Saturday and Sunday. In Beijing time the peak windows are 09:00–12:00 and 14:00–18:00 on weekdays: DeepSeek charges double during its home market’s working hours and half everywhere else.
What a month costs
Each workload is 80% input and 20% output. Sorted by cost.
| Model | 10M tokens, off-peak | 10M tokens, peak | 100M tokens, off-peak | Off-peak, 50% of input cached |
|---|---|---|---|---|
| deepseek-flash | $2.40 | $4.80 | $24.00 | $1.81 (10M) |
| deepseek-v4-pro | $9.24 | $18.48 | $92.40 | $6.69 (10M) |
There is no Batch API discount on this price list. Off-peak is the batch discount: work that can wait for the evening or the weekend costs exactly half, with no separate API to integrate.
So, is it the cheapest?
Two different questions hide inside that one.
Cheapest tokens, outright: no. GPT-6 Luna and Claude Haiku 5.5 list at $0.10 input / $0.50 output — below deepseek-flash’s $0.15 / $0.60 off-peak, and far below its peak rates. At an 80/20 mix, Luna costs $1.80 per 10M tokens against Flash’s $2.40 off-peak and $4.80 peak. If your workload is simple and price is the only variable, DeepSeek is not the floor.
Cheapest flagship-class reasoning: yes, and it is not close. deepseek-v4-pro at $0.66 / $1.98 off-peak undercuts every closed flagship on list price. At the same 80/20 mix, V4 Pro costs $9.24 per 10M tokens off-peak and $18.48 at peak — against $36 for Claude Sonnet 5.5, GPT-6.1 Sol or Gemini 4 Argon at introductory pricing, and $180 for GPT-6 Astra. Even DeepSeek’s peak rate is half the standard rate of the cheapest closed flagship tier.
Cache behaviour is where DeepSeek claws back the token gap. A cache hit on Flash costs $0.003 per 1M — 2% of its cache-miss price, and below the $0.01 cached rate of Luna and Haiku. With half of your input served from cache, Flash off-peak lands at $1.81 per 10M tokens, level with Luna’s standard-rate bill.
Billing rules that change the bill
- The clock is a pricing tier. The same request costs twice as much at 09:30 UTC on a Tuesday as at 15:00 UTC or any time on a weekend. Schedulers and queues should treat the peak windows (01:00–04:00 and 06:00–10:00 UTC, weekdays) as a 2x zone.
- Cache hits are nearly free, and automatic. Input served from DeepSeek’s context cache costs 2% of the miss price on Flash ($0.003 vs $0.15) and about 3% on V4 Pro ($0.022 vs $0.66). A stable system prompt or document prefix is worth more here than on any other major API.
- Old model names silently map to the new Flash. The retired names
deepseek-v4-flashanddeepseek-v4-flash-vision-expstill work, but requests are served by DeepSeek-V4.1-Flash and billed at the Flash price. Budgets keyed to those names are Flash budgets now. - R1 is no longer on the price list. The current official list covers the two V4 models only. Anyone still budgeting around R1-era rates is working from a price list DeepSeek no longer publishes.
- Agent tool calls add a per-request line from November 1, 2026. DeepSeek’s standalone Search and Fetch APIs cost $0.75 per 1,000 successful requests, and server-side tool calls made through agent harnesses start being billed at that rate from November 1, 2026 — a small but real addition for coding-agent workloads.
Which model to price first
- deepseek-flash: the default. Vision included, the highest concurrency limit (2,500), and the cheapest cache hits in the industry. Start here unless you have evidence Pro changes your results.
- deepseek-v4-pro: the reasoning flagship — text-only, concurrency capped at 500, and output at 3.3x the Flash price in both peak and off-peak. Worth it for hard reasoning and coding; measurable, not automatic.
Price your own traffic on the DeepSeek V4 Pro page, see the full flagship comparison in DeepSeek V4 Pro vs closed models, or browse every official DeepSeek model on the DeepSeek provider page.
Prices: DeepSeek API pricing, verified 2026-10-26. Prices change; check the official page before committing budget.