DeepSeek API pricing in 2026 — what you actually pay

DeepSeek's official API charges $0.15 per million input tokens and $0.60 per million output tokens for its Flash model — off-peak. In peak hours the price doubles, and input served from the cache costs fifty times less. Those two factors move the bill more than the headline price does. Below: the current price list, how peak hours and the cache work, two worked examples and the cases where a fixed price comes out cheaper.

Official DeepSeek API prices#

Prices per 1M tokens from DeepSeek's pricing page, checked on October 20, 2026:

Both models take up to 1M tokens of context and return up to 384K output tokens. DeepSeek reserves the right to change prices, so check the page before you plan a budget.

Peak hours are 01:00–04:00 and 06:00–10:00 UTC on weekdays, except Chinese public holidays. Weekends are off-peak all day.

The cache is automatic. If the beginning of a request fully matches a prefix DeepSeek has already stored — a system prompt, tool descriptions, earlier chat history — that part is billed at the cache-hit price. It works on a best-effort basis: there's no guaranteed hit rate, and unused cache is cleared within hours to days.

How the bill adds up#

For every request:

cost = cache_hit_tokens  / 1,000,000 × hit_price
     + cache_miss_tokens / 1,000,000 × miss_price
     + output_tokens     / 1,000,000 × output_price

You don't have to guess the split: DeepSeek returns prompt_cache_hit_tokens and prompt_cache_miss_tokens in the usage block of every response.

Example 1: a chatbot, 10,000 requests a month#

Each request sends 2,000 tokens and gets a 500-token answer. That's 20M input and 5M output tokens a month.

Output is a large share of this bill, and output is where the gap is widest: $0.60–1.20 per million at DeepSeek versus $0.10. Even a perfect cache can't close it, because the cache only discounts input.

Example 2: a coding agent, 100 tasks a month#

An agent resends the whole conversation on every step, so its bill is mostly input. Take a 30-step task with 1.05M input and 15K output tokens (we walk through this calculation in how much an AI agent costs) and run 100 such tasks a month: 105M input and 1.5M output tokens.

Here the cache decides everything: the more of the agent's input DeepSeek serves from the cache, the closer the official price gets to the fixed one. The CheapTokens bar always shows the current price; a 5% fee is withheld on top-ups, so add it when you compare. The official API can come out cheaper only off-peak and with a high, stable cache hit rate. With fewer cache hits, or in peak hours, the fixed price wins.

When a fixed price pays off#

CheapTokens sells DeepSeek-V4-Flash at $0.0221 per 1M input tokens and $0.10 per 1M output tokens, with no peak hours and no separate cache price. It's the 0731 release, not the V4.1-Flash-1017 that deepseek-flash points to today, so compare answer quality on your own tasks, not only the price.

A fixed price is the better deal when:

• answers are long relative to the prompt — chat, generation, translation;
• requests are short and varied, so the cache rarely hits;
• your traffic falls into DeepSeek's peak hours;
• you'd rather pay in crypto or by card through a partner, and keep one balance for DeepSeek, GLM and MiniMax.

The official API can win on long agent sessions outside peak hours, when a stable prefix keeps the cache hit rate high — check your own share of cache hits against the chart above. It's also the choice when you specifically need V4.1-Flash or V4-Pro.

Switching is a two-line change: the API is OpenAI-compatible, so you replace the base URL and the key and keep the rest of your code. Top-ups start at $5 in crypto or $10 by card through a partner. A 5% fee is withheld on each top-up: to cover $1.00 of usage, you add about $1.05 to the balance.

Check it on your own traffic#

Take a week of real requests and add up three numbers: input tokens, output tokens and, for DeepSeek, the share of cache hits. Multiply by the prices above and you'll know which option is cheaper for your workload. Current CheapTokens prices: