DeepSeek API cost calculator
Headline prices are quoted per million tokens at a single rate, which is not how DeepSeek actually bills. Input is metered twice depending on cache hits, and the whole card moves with the time of day. This calculator applies all four rates to your own volumes and shows what the same workload would cost on Anthropic, OpenAI and Google.
Share of input tokens DeepSeek can serve from cache. Coding assistants and agent loops re-send the same prefix constantly and often sit above 90%.
Drag to zero to see what a fully off-peak schedule would cost.
From 16 Aug 2026
—
per month · — per day
Before the change
—
per month · — per day
Running the same volume entirely in off-peak hours: — per month, saving —.
Same volume, other providers · monthly
- DeepSeek V4-Pro —
- Anthropic Claude Sonnet 5 — —
- Anthropic Claude Opus 5 — —
- OpenAI GPT-5.6-sol — —
- OpenAI GPT-5.6-terra — —
- OpenAI GPT-5.6-luna — —
- Google Gemini 3.1 Pro Preview — —
Comparisons assume the same token volumes and the same cache hit rate everywhere, which is a simplification: cache behaviour differs by provider. Only DeepSeek charges by time of day, so the other rows are flat.
Per-provider caveats
- Anthropic Claude Sonnet 5 — Writing to the 5-minute cache costs $2.50 per million tokens on top of the read rate.
- Anthropic Claude Opus 5 — Anthropic charges a separate premium for writing to the cache.
- OpenAI GPT-5.6-sol — Caching is automatic; there is no separate cache-write charge.
- OpenAI GPT-5.6-terra — Caching is automatic; there is no separate cache-write charge.
- OpenAI GPT-5.6-luna — Caching is automatic; there is no separate cache-write charge.
- Google Gemini 3.1 Pro Preview — Rates shown apply to prompts of 200K tokens or fewer; longer prompts move to a higher tier. Google charges hourly storage for explicit context caches on top of the read rate.
How the estimate is built
Input is split before it is priced
Your daily input volume is divided by the cache hit rate, then each half is charged at its own rate. At the default 80% hit rate, most of your input bills at the cheaper of the two, which is why a single "input price" is misleading.
The day is split too
The peak-hour slider blends the two rate cards. It is a proportion of traffic, not a schedule — set it to match how much of your volume genuinely lands inside the peak windows.
A month is 30 days
Monthly figures are the daily cost times 30. Real months run 28–31 days, so treat the monthly number as a planning figure rather than an invoice.
Comparisons are deliberately simple
Other providers are priced at the same token volumes and the same cache hit rate. None of them charge by time of day, and each handles cache writes differently — the comparison shows the shape of the gap, not a contractual quote.
What the profiles assume
| Profile | Input/day | Output/day | Cache hits |
|---|---|---|---|
| Light Occasional chat and one-off scripts. A few dozen calls a day. | 200K | 40K | 40% |
| Daily coding A coding assistant open all day, repeated turns over the same files. | 5M | 400K | 85% |
| Heavy agent Long autonomous runs, large re-read context, high tool-call volume. | 40M | 3M | 90% |
Profiles are starting points for DeepSeek V4-Pro, not measurements. Replace them with figures from your own usage dashboard whenever you have them.
Calculator questions
How much cheaper is the DeepSeek off-peak rate?
Off-peak rates are 50% lower than peak, applied to input and output alike. On DeepSeek V4-Pro that is $1.98 per million output tokens instead of $3.96.
It is worth being clear about what the discount is measured against: it is half of the new peak rate, not a return to the old one. Off-peak output still costs 2.3× what the same tokens cost under the previous flat card.
The rate is decided by when the request reaches DeepSeek, not when your job was queued locally. Batch work that can tolerate delay is the clearest way to capture it.
How does cache-hit pricing work?
Input tokens are billed at two different rates. Tokens DeepSeek can serve from its prompt cache cost $0.044 per million at peak; tokens it has to process fresh cost $1.32 per million — a gap of 30×.
Caching keys on identical prefixes, so it rewards keeping the stable part of a prompt — system instructions, tool definitions, a file you are iterating on — byte-identical at the front, with the varying part at the end. Reordering that prefix between calls throws the cache away.
Coding assistants and long agent loops re-send the same context repeatedly and commonly sit above 90% cache hits. That is why the calculator on this site defaults to 80% rather than assuming every token is billed at full price.
What is the difference between V4-Pro and V4-Flash?
Both carry a 1M token context window and the same maximum output length. They differ on price and on how many requests you may run at once: DeepSeek V4-Pro allows 500 concurrent requests, DeepSeek V4-Flash allows 2,500.
DeepSeek V4-Flash output costs $1.32 per million at peak against $3.96 for DeepSeek V4-Pro. Both models moved to the new card on the same date and share the same peak windows.
Is DeepSeek still cheaper than Claude, OpenAI and Gemini after the increase?
For most workloads, yes — but by a smaller margin than before, and the gap narrows further against the cheaper tiers of each provider once you account for cache hits.
The right comparison depends on your own mix of input, output and cache rate, which is what the calculator is for: it applies your volumes to every rate card at once instead of comparing headline numbers that assume a workload you may not have.