How much does DeepSeek API cost?
DeepSeek's API is pay-as-you-go and very cheap. As of 2026-07, the flash model (deepseek-chat / v4-flash) costs about $0.14 per million input tokens and $0.28 per million output tokens; the pro reasoning model (deepseek-reasoner / v4-pro) about $0.435 in and $0.87 out. Cache hits cost a tiny fraction, and there's no monthly subscription.
Why — the first-principles explanation
You don't pay DeepSeek a monthly fee; you pay per token, where a token is roughly three-quarters of a word. Two things get metered separately: input tokens (the text you send, including your prompt and history) and output tokens (the text the model writes back). Output is priced higher because generating new text takes far more computation than reading existing text.
A second lever is context caching. If parts of your prompt repeat — a long system instruction, tool definitions, a document you keep referencing — DeepSeek stores that processed text and charges a cache hit rate that's roughly a fiftieth of the normal input price. This is why the same conversation can cost dramatically less on the second and later turns, and why well-designed apps put stable content at the front of the prompt.
The headline is how cheap this is: at around fourteen cents per million input tokens, DeepSeek is often ten to thirty times less expensive than comparable Western frontier models. That price gap comes from DeepSeek's efficiency-first engineering (Mixture-of-Experts, memory-saving attention), which lets each answer burn less compute — savings it passes on to win developers.
An example that makes it click
Think of it like a photocopier that bills by the page. Reading pages you feed in (input) is cheap; printing new pages (output) costs a bit more. A million pages fed in costs about fourteen cents, and a million printed comes to about twenty-eight cents — pocket change for most everyday use.
And there's a loyalty trick: if you keep copying the same cover letter over and over, the machine remembers it and charges you almost nothing for that repeated page (a cache hit). So a chatbot that sends the same instructions every message pays full price once, then a tiny fraction after that. For a hobby project, a whole month of use might cost less than a cup of coffee.
Key facts
- Billing is pay-as-you-go per token; there is no monthly subscription for API access.
- deepseek-chat (v4-flash): ~$0.14 per 1M input tokens (cache miss), ~$0.28 per 1M output tokens (as of 2026-07).
- deepseek-reasoner (v4-pro): ~$0.435 per 1M input tokens and ~$0.87 per 1M output tokens (promotional; list rates are higher).
- Cache-hit input is far cheaper — roughly $0.0028 per 1M tokens for the flash model — when prompt prefixes repeat.
- This is often 10-30x cheaper than comparable frontier models; a token is roughly 0.75 words.
An open-weight Chinese model family that matched frontier quality at low cost.
Affiliate link — we may earn a commission at no cost to you.▶ The 60-second explainer (script)
How much does the DeepSeek API cost? Very little. There's no subscription — you pay per token, which is about three-quarters of a word. As of mid-2026, the fast flash model costs roughly fourteen cents per million input tokens and twenty-eight cents per million output tokens. The more powerful reasoning model is a bit more, around forty-four cents in and eighty-seven cents out per million. And here's the money-saver: if parts of your prompt repeat, cached input costs a tiny fraction — about a fiftieth of the normal price. Compared to other top models, DeepSeek is often ten to thirty times cheaper, because its efficient design burns less compute per answer. For most hobby projects, a whole month of usage can cost less than a cup of coffee.
What authoritative sources say
People also ask
Is there a subscription fee?
No. The API is pure pay-as-you-go — you're billed only for the tokens you use.
Which model is cheaper?
The flash model (deepseek-chat / v4-flash) is cheapest; the reasoning model (deepseek-reasoner / v4-pro) costs more per token but is better at hard logic.
How can I lower my API bill?
Reuse stable prompt prefixes to trigger cache hits, keep prompts concise, and pick the flash model when reasoning isn't needed.
Are prices likely to change?
Yes — DeepSeek periodically updates pricing and runs promotions, so check the official pricing page for current rates.