DeepSeek pricing: there is no monthly plan, and off-peak costs half
DeepSeek has no subscription. Web chat is free and everything else is per token, with off-peak rates at half the peak price. V4-Flash runs $0.22 per million input tokens off-peak.
Verdict
DeepSeek sells no monthly plan. Web chat is free and the API is per token, with off-peak rates half of peak. V4-Flash is $0.22 per million input tokens off-peak against $0.44 at peak.
There is no plan to buy#
No subscription
DeepSeek's consumer pricing
Free chat, then per-token API. Nothing in between.
Every other assistant in this comparison sells you a monthly plan. ChatGPT has seven tiers, Claude has seven, Google has four.
DeepSeek has none. The structure is:
- Web chat: free. No paid tier above it
- API: per token. No monthly commitment
That is the whole product. If you are a consumer, you pay nothing. If you are a developer, you pay for exactly what you use.
The API rates#
Effective 16 August 2026, per million tokens:
| Model | Peak | Off-peak | |
|---|---|---|---|
| V4-Flash | Input | $0.44 | $0.22 |
| Output | $1.32 | $0.66 | |
| V4-Pro | Input | $1.32 | $0.66 |
| Output | $3.96 | $1.98 |
Off-peak is exactly half of peak on every line. That is unusual and it is the most useful thing on this page.
The peak windows are published in UTC and are narrow. Because they are stated in UTC rather than your local time, check DeepSeek's own pricing page and translate them to your working day before you build a schedule around them.
The gap against the frontier models#
This is the comparison that makes DeepSeek worth knowing about:
| Per million tokens | Input | Output |
|---|---|---|
| GPT-6 Astra | $10.00 | $50.00 |
| DeepSeek V4-Pro, off-peak | $0.66 | $1.98 |
| DeepSeek V4-Flash, off-peak | $0.22 | $0.66 |
Against GPT-6 Astra, V4-Flash off-peak is roughly 45x cheaper on input and 75x cheaper on output.
These are not equivalent models and we are not claiming they are. GPT-6 Astra has a 1,050,000-token context window and frontier reasoning. But the gap is large enough that "we default to the best model" is a decision worth pricing rather than assuming.
What no subscription means for a small business#
The absence of a plan cuts both ways.
Good: nothing to cancel, no seat you forgot about, no annual commitment. If you use DeepSeek's chat occasionally, it costs you nothing, ever.
Less good: if you build a product on the API, you have a usage bill, not a budget line. A subscription is predictable. Per-token billing means a busy month costs more than a quiet one, and a bug in a retry loop costs more than either.
If you go this route, set a spend alert on day one. That is true of any usage-billed API and it is truest of the cheap ones, because the low unit price makes it easy to stop paying attention.
The cache rate we could not settle#
Sources report DeepSeek's cached input at $0.007 and at $0.028 per million tokens, both recently, for what appear to be different model builds.
We are not going to pick one. The practical advice is the same either way: budget from the standard input rate, treat caching as a saving you discover rather than one you plan on, and confirm the current figure on DeepSeek's own page if the difference matters to your forecast.
Who this is actually for#
| Your situation | Verdict |
|---|---|
| You want a free assistant with no account admin | Use the web chat. It costs nothing |
| You are building and cost is the constraint | Strongest option here by a wide margin |
| You have schedulable batch work | Off-peak halves it. Do this |
| You need frontier reasoning on hard problems | Test against a frontier model first |
| You want a predictable monthly bill | This is the wrong product. Buy a subscription elsewhere |
For a small business that just wants to ask an AI questions, the free chat is genuinely free and there is no upsell waiting. That is rare enough to be worth saying plainly.
The short version
What works
- Web chat is free with no plan to cancel, which is not true of any other major assistant
- Off-peak API rates are half the peak rate, and off-peak covers most of the week
- Even at peak, the API is one to two orders of magnitude cheaper than the frontier US models
What does not
- No subscription means no predictable monthly bill if you build on the API, only a usage bill
- Peak and off-peak windows are defined in UTC, so you have to translate them to your own working day
- Cached input pricing is reported inconsistently across sources, so budget from the standard rate rather than the cache rate
Frequently asked questions
- Is DeepSeek free?
- The web chat is free to use, yes, and there is no paid tier above it to upgrade to. What costs money is the API, which is billed per token with no monthly subscription attached. That makes DeepSeek the only major assistant in this comparison with no subscription product at all.
- How much does the DeepSeek API cost?
- As of 16 August 2026, V4-Flash is $0.44 per million input tokens and $1.32 output at peak, and $0.22 input and $0.66 output off-peak. V4-Pro is $1.32 input and $3.96 output at peak, and $0.66 and $1.98 off-peak. Off-peak is half of peak across the board.
- When are DeepSeek off-peak hours?
- Sources describe peak windows in UTC, with off-peak covering most of the rest of the week. Because the definition is given in UTC and the windows are narrow, confirm the current schedule on DeepSeek's own pricing page before you build a scheduler around it.
- How does DeepSeek compare to GPT-6 Astra on cost?
- GPT-6 Astra is $10 per million input tokens and $50 output. DeepSeek V4-Flash off-peak is $0.22 and $0.66. That is roughly 45 times cheaper on input and 75 times cheaper on output. They are not equivalent models, but the gap is large enough that the comparison is worth making before defaulting to a frontier model.
Written by
Tashawar Awais
Researcher and editor
Tashawar handles verification and editing. Every figure in a review is checked a second time before it goes out, and anything that cannot be traced to a vendor page or a documented source is either qualified or cut. Where pricing is genuinely unclear, as it is with Canva team plans or Close CRM tiers, the article says so and tells the reader to confirm directly instead of quoting a number with false confidence.
- Second check on every published figure
- Removes claims the sources do not support
- Flags pricing that changes without notice