DeepSeek
A free assistant with open-weight models and per-token API pricing an order of magnitude below the field.
- AI
- Development
- from $0.22
DeepSeek is a Chinese AI lab whose assistant is free to use with no paid consumer tier, and whose models are published as open weights. Its commercial product is the API, priced per token at a fraction of the Western labs, with off-peak discounts. It publishes no business tier, admin console or single sign-on.
DeepSeek: the short answer
Every line below is drawn from facts verified on 19 August 2026.
- What it is
- A free assistant with open-weight models and per-token API pricing an order of magnitude below the field
- Best for
- Individuals who want a capable assistant at no cost
- Starts at
- Free; API from $0.22 per million input tokens
- Free option
- Free plan available, no trial
- Biggest strength
- The assistant is free, with no paid tier and nothing withheld behind one
- Biggest drawback
- No business tier, admin console, SSO or audit logs published
- Based on
- Primary-source research, not hands-on use
- Closest alternative
- Mistral - Also open-weight, with a published business tier, EU terms and a data processing addendum
DeepSeek at a glance
| Best for | Individuals who want a capable assistant at no cost, Developers building on a very cheap API, Teams who want to run open weights on their own hardware |
|---|---|
| Category | AI, Development |
| Starting price | Free; API from $0.22 per million input tokens |
| Free plan | Yes |
| Free trial | No trial offered |
| Pricing model | usage based |
| Company | DeepSeek |
| Founded | Not disclosed |
| Headquarters | China |
| Founders | Not disclosed |
| Platforms | Web, iOS, Android |
| Public API | Yes |
| Browser extension | No |
| Mobile apps | iOS + Android |
| Self-hosting | Yes |
| Affiliate programme | No |
| Data source | Vendor pricing page and public records |
| Last checked | August 2026 |
DeepSeek pricing
Prices below were read from the vendor's own pricing page on 19 August 2026. Pricing model: usage based. No free trial is offered.
| Plan | Price | Billing | Notes |
|---|---|---|---|
| Assistant | $0 | per month | Free on the web and in the mobile apps, with no paid consumer tier published. No usage figures are published either |
| API, deepseek-v4-flash | $0.22 | per usage | Per million input tokens on a cache miss, off-peak; $0.44 at peak. Cache hits are $0.007 off-peak. Output is $0.66 off-peak and $1.32 at peak |
| API, deepseek-v4-pro | $0.66 | per usage | Per million input tokens on a cache miss, off-peak; $1.32 at peak. Cache hits are $0.022 off-peak. Output is $1.98 off-peak and $3.96 at peak |
Which DeepSeek plan should you buy?
| Your situation | Recommended | Why |
|---|---|---|
| You want a capable assistant and will not pay | The free assistant | There is no paid consumer tier, so the free version is the product rather than a sample. |
| You are building on an API at volume | deepseek-v4-flash | From $0.22 per million input tokens off-peak on a cache miss, and $0.007 on a cache hit. Schedule batch work off-peak. |
| You need the stronger model for harder tasks | deepseek-v4-pro | $0.66 per million input tokens off-peak, still far below the field. |
| You need the model inside your own infrastructure | Self-host the open weights | The weights are published, which is the only route here that removes the vendor entirely. |
| You are buying for a company | Do not buy DeepSeek on the published information | No admin console, SSO or audit logs are published, and nothing on the site addresses data residency. |
Verdict by who you are
Individual on a tight budgetYes
A free assistant with no paid tier withholding features is a genuine offer, not a funnel.
Developer with high API volumeYes
Per-token pricing an order of magnitude below the field, with caching and off-peak rates on top.
IT team rolling out to a companyNo
No published admin console, SSO or audit logging, and no published data-processing position.
Regulated industry or public sectorNo
Data residency cannot be established from the vendor's own pages, which is usually the end of the conversation.
Team wanting open weights with enterprise termsCompare first
Mistral offers open weights alongside a published DPA, EU terms and a priced business tier.
Should you buy DeepSeek?
Buy DeepSeek if…
- Individuals who want a capable assistant at no cost
- Developers building on a very cheap API
- Teams who want to run open weights on their own hardware
Do not buy DeepSeek if…
- Company rollouts needing admin controls, SSO or audit logs
- Regulated industries, or anyone with data-residency obligations
- Buyers who need a vendor that publishes its corporate details
Best DeepSeek alternative for each job
Chosen by what you are trying to do rather than by which product is closest overall, because the right replacement depends on the job.
| If you mainly need to… | Use | Why |
|---|---|---|
| Open weights with enterprise paperwork | Mistral | Open-weight models plus a published data processing addendum and a $24.99 team seat. |
| Roll an assistant out to staff | Microsoft Copilot | From $18 a user a month paid yearly, with the admin controls DeepSeek does not publish. |
| Long-document analysis | Claude | What it is bought for, at $200 a year on annual billing. |
| Cited answers for research | Perplexity | Sources you can open, at $20 a month. |
Key features
- Free assistant on web and mobile with no paid tier to upsell
- Open-weight models published for download
- API priced per million tokens, with peak and off-peak rates
- Context caching, which cuts input cost by roughly 97% on a cache hit
- Reasoning models available at the same low rates
- Models runnable on your own hardware
- Published model releases on GitHub
- Cheap enough to make high-volume batch work practical
Integrations
- DeepSeek API, with an OpenAI-compatible interface
- Open weights distributed through GitHub and model hosts
- Self-hosted deployment
DeepSeek screenshots
Captured by us in August 2026 at 1440px, unedited. Interfaces change, so these show the product as it looked on the date we verified this profile. Pricing pages are captured from a US vantage point, so the currency shown matches the USD figures recorded above.
Our assessment
DeepSeek is the cheapest serious option in this category by a wide margin, and the reason is structural rather than promotional: the assistant is free with no paid tier, the models are published as open weights, and the API is priced per token at roughly a tenth of what the Western labs charge, with off-peak rates and context caching cutting it further. For an individual, or for a developer moving high-volume work off an expensive API, that is a real and defensible saving. For a company, it is the hardest product here to approve, and the reasons are not about capability. There is no published business tier, no admin console, no single sign-on and no audit logging visible on its own pages. The company publishes very little about itself - no founding date, no offices, no funding - and its site carries Chinese provincial filings, which makes data residency and processing a question you must answer with your own legal advice rather than from a vendor page. Our honest position: excellent value for personal use and self-hosted work, and not a purchase we would put in front of a procurement process on the strength of what is published.
What works in DeepSeek
- The assistant is free, with no paid tier and nothing withheld behind one
- Open weights make self-hosting and a genuine exit route possible
- API pricing roughly an order of magnitude below the Western labs
- Context caching and off-peak rates cut real costs further still
What doesn't
- No business tier, admin console, SSO or audit logs published
- No published usage figures for the free assistant
- The company publishes almost no corporate detail, and none about data residency
- Narrow next to the general assistants, with no published image, voice or video generation
EveryAny Decision Score
3 of these 8 dimensions are calculated from verified facts; the rest are our judgement, shown as a band rather than a figure unless something specific sits behind them. How each score works.
- Cost position10/10
derivedno paid tier required
- Portability8/10
derivedself-hostable, public API
- Subscription flexibility6/10
derivedfree plan, usage based, monthly billing
- Feature depth5/10
editorialAn assistant and an API. There is none of the product surface - projects, connectors, agents - that the priced assistants build around the model.
- Ease of learningHigh
editorialour judgement, with no measurement behind it
- Collaboration1/10
editorialNo teams, no shared workspace, no admin console and no SSO published; it is a single-user assistant and an API.
- AI capability6/10
editorialText and reasoning models, published as open weights. No image, voice or video generation is advertised on its own pages.
- Professional adoptionHigh
editorialour judgement, with no measurement behind it
What leaving DeepSeek would cost
low lock-in
DeepSeek scores on self-hostable, public API. Portability 8/10 ; lock-in is the inverse. How this is calculated.
DeepSeek competitors and alternatives
| Alternative | Why you might pick it instead |
|---|---|
| Mistral | Also open-weight, with a published business tier, EU terms and a data processing addendum. |
| ChatGPT | Far more product around the model, with a business route and admin controls. |
| Claude | The usual choice where long documents and careful instruction-following matter. |
ChatGPT
AI
General-purpose AI assistant for writing, analysis, coding and research.
Checked Aug 2026
Mistral
AI
A European assistant with open-weight models, publishing limits as actual numbers.
Checked Aug 2026
The API price is the headline, and it is not close
Per million tokens, read from DeepSeek’s own documentation on 19 August 2026:
| Cache hit, input | Cache miss, input | Output | |
|---|---|---|---|
| deepseek-v4-flash, off-peak | $0.007 | $0.22 | $0.66 |
| deepseek-v4-flash, peak | $0.014 | $0.44 | $1.32 |
| deepseek-v4-pro, off-peak | $0.022 | $0.66 | $1.98 |
| deepseek-v4-pro, peak | $0.044 | $1.32 | $3.96 |
Two mechanics are worth building around. Context caching cuts input cost by roughly 97% on a hit, which rewards prompts with a stable prefix. And off-peak pricing halves the rate, which rewards scheduling batch work rather than running it on demand.
The assistant is free, and that is the whole tier list
There is no DeepSeek Plus, Pro or Max. The web and mobile assistant is free, and no paid consumer tier is published, which makes it the only product in our AI assistants guide where the free version is not a sample of a paid one.
It also publishes no usage figures at all - but then almost nobody in this category does, so it loses less by comparison than it might.
Why we would not put it through procurement
This is the part a price comparison misses. On DeepSeek’s own pages there is:
- No business or team tier
- No admin console, provisioning or single sign-on
- No audit logging
- No published data-residency or processing position, and no data processing addendum
Its site carries Chinese provincial filings and very little else about the company - no founding date, no offices, no funding. None of that makes the model worse. All of it makes the compliance conversation one you will have with your own lawyers rather than with a vendor page, and for regulated industries that is usually where it ends.
If open weights are what attracts you and the paperwork matters, Mistral publishes both: open-weight models and a data processing addendum, with a team seat at $24.99.
Where it is genuinely the right answer
Two cases, both strong. An individual who wants a capable assistant and will not pay for one - the free tier is the product, not a funnel. And a developer with real volume, where an order-of-magnitude cheaper API, caching and off-peak scheduling turn an unaffordable feature into an affordable one. Neither case needs an admin console.
DeepSeek questions, answered
- Is DeepSeek worth it?
- DeepSeek is the cheapest serious option in this category by a wide margin, and the reason is structural rather than promotional: the assistant is free with no paid tier, the models are published as open weights, and the
- How much does DeepSeek cost?
- Free; API from $0.22 per million input tokens. Verified against the vendor's pricing page.
- Does DeepSeek have a free plan or free trial?
- DeepSeek has a free plan. No free trial is offered.
- What is the best DeepSeek alternative?
- It depends on the job. Mistral: Also open-weight, with a published business tier, EU terms and a data processing addendum. ChatGPT: Far more product around the model, with a business route and admin controls. Claude: The usual choice where long documents and careful instruction-following matter.
- Can I self-host DeepSeek?
- Yes. DeepSeek can be run on your own infrastructure, so the data stays under your control.
What this profile is based on
Every claim below is about our work, not about DeepSeek. A "no" is recorded rather than omitted, because the difference between a tested review and careful desk research is the reader's to weigh. How we research a profile.
-
Prices read from the vendor : yes
Read from DeepSeek's own pricing page on 19 August 2026
-
Plans and terms recorded : yes
3 rows, each with its billing period
-
Original screenshots : yes
1 captured by us: pricing page
-
Sources cited and dated : yes
2 primary sources, each with the date it was read
-
Product used first-hand : no
Not tested. This profile is desk research from primary sources
-
Alternatives used first-hand : no
None, so comparisons here rest on published facts rather than use
-
Vendor contacted : no
Not contacted for this profile
Desk research from DeepSeek's own site and API documentation, read on 19 August 2026 through a United States egress. Nobody here has used DeepSeek, and no model quality was benchmarked. Its site publishes very little corporate information; what is recorded as "China" comes from the provincial filings in its own footer, and we make no claim beyond that.
One-click Google setting
Get our writing surfaced first on Google
If this was useful, add EveryAny.One as a preferred source in Google. Our profiles and best-of guides will appear ahead of the rest in Top Stories, AI Mode and AI Overviews when you search for software.
- Our reviews get a "preferred" badge in Top Stories and AI Overviews.
- AI Mode leans on the sources you've chosen when it answers.
- Change it back any time from the same Google page.
Opens Google's source preferences in a new tab.
Provenance and freshness
When each fact was last checked
- Pricing ·
- Plans ·
- Free trial
- Features
- Platforms
- Integrations
- Affiliate programme
- Company details ·
Individually sourced claims
-
deepseek-v4-flash costs $0.22 per million input tokens on a cache miss off-peak, and $0.007 on a cache hit
DeepSeek API pricing ↗ · checked
-
deepseek-v4-pro costs $0.66 per million input tokens on a cache miss off-peak, with output at $1.98
DeepSeek API pricing ↗ · checked
-
DeepSeek publishes no paid consumer tier for its assistant
DeepSeek ↗ · checked
What has changed
-
August 2026
- First published, with API rates read from the vendor's own documentation
- Recorded the absence of a business tier, admin controls and corporate detail
Sources & verification
Facts on this page were read from the sources below on the dates shown. Where a vendor does not publish a figure, we record it as not disclosed rather than estimating it.
- DeepSeek - checked 19 August 2026
- DeepSeek API pricing - checked 19 August 2026
Spotted something out of date? Email [email protected] and we will re-check it. Corrections are logged on the page.
