Yes, the DeepSeek chat app (web at chat.deepseek.com, plus iOS and Android) is free, with no subscription tier, no ads, and no published hard daily message cap, only fair-use throttling during heavy traffic. It routes everyday questions to DeepSeek-V4-Flash and harder reasoning tasks to DeepSeek-V4-Pro, both free. The API, a separate, metered product for developers, is billed per token and just got dramatically pricier: up to 1,100% at peak hours as of 16 August 2026. The price hike is developer-facing only. It did not touch the free chat app.
The bigger caveat isn't cost. DeepSeek stores data in China under a law that can compel disclosure to Chinese authorities, and several countries have banned or restricted it as a result. That's worth weighing regardless of what you pay.
"Is DeepSeek free" gets asked constantly because the answer genuinely depends on which DeepSeek you mean, and DeepSeek itself doesn't make that obvious on the homepage. This breaks down exactly what costs nothing, what's metered, how the mid-August price change actually affects you, and where people get burned.
What's actually free
- The chat app, web and mobile, no subscription tier exists to upgrade to. Signing in syncs your conversations across web, iOS, and Android.
- Both current models are available to free users: DeepSeek-V4-Flash (284B total, 13B active parameters) handles everyday questions in "Instant" mode; DeepSeek-V4-Pro (1.6T total, 49B active parameters) handles complex reasoning in "Expert" mode.
- No published hard daily cap. The only real constraint is fair-use throttling during high-traffic periods, you'll see slower responses or a "Server Busy" message, not a paywall.
- R1's open weights, downloadable free from Hugging Face under an MIT license, for anyone willing to self-host on their own hardware. This is the one route that's free forever, independent of DeepSeek's own servers.
What costs money: the API
The free chat app and DeepSeek's API are fully separate products with separate billing. Building anything that calls DeepSeek's models programmatically, a script, an app, an integration, gets billed per token through platform.deepseek.com regardless of how much you personally use the free chat app.
As of the 16 August 2026 price change, DeepSeek bills on a peak/off-peak schedule (peak = 01:00-04:00 and 06:00-10:00 UTC; everything else is off-peak):
Model Cache-hit input Cache-miss input Output Concurrency
($/M tokens) ($/M tokens) ($/M tokens) limit
-----------------------------------------------------------------------------------
DeepSeek-V4-Flash $0.007 / $0.014 $0.22 / $0.44 $0.66 / $1.32 ~2,500
DeepSeek-V4-Pro $0.022 / $0.044 $0.66 / $1.32 $1.98 / $3.96 ~500
(off-peak/peak) (off-peak/peak) (off-peak/peak)
Source: api-docs.deepseek.com/quick_start/pricing, verified 19 Aug 2026That's the same hike covered in our DeepSeek V4 Pro price-hike article, landing here because it's the fact that answers the "is it free" question for anyone building with the API rather than just chatting. The reasoning-specific API model, once called deepseek-reasoner (the R1-era alias), was retired 24 July 2026; reasoning is now built into V4 itself, both Flash and Pro run "thinking mode" by default. That default matters for cost: a short, two-line visible answer can silently generate thousands of billed reasoning tokens before you see the final reply, which is the single most common cause of a bigger-than-expected API bill.
New API accounts commonly report a starting credit around 5 million free tokens (roughly $8-9 of notional value), valid for about 30 days. That figure isn't documented as a fixed, guaranteed policy in DeepSeek's own docs, so treat it as something that may appear in your dashboard rather than an advertised, permanent free tier.
Did the August price hike change the free app too?
No. Every source covering the 13-17 August 2026 launch of V4 Pro and the subsequent price increase, DeepSeek's own announcement, Fortune, Engadget, and others, scopes the change entirely to developer/API pricing. DeepSeek's stated rationale was shifting developer load toward off-peak hours to manage server capacity; nothing in the announcement or the coverage that followed mentions any change to the free consumer app's model access or usage limits. If you only use the chat app, the price hike doesn't touch you at all.
What the API actually costs in practice
The per-token rates above are hard to feel until you attach them to a real task. Take a common one: summarizing 100 documents averaging 5,000 words (roughly 6,500 tokens) each, using DeepSeek-V4-Flash off-peak, with cache misses (a fresh document each time, no repeated context to cache). Input cost: 100 × 6,500 tokens × $0.22/M ≈ $0.14. Output, assuming a 500-token summary per document: 100 × 500 tokens × $0.66/M ≈ $0.03. Total: well under $1 for the batch. Run the same job at peak hours instead of off-peak and the cost roughly doubles, still under $2. The API is genuinely cheap at Flash-tier volume; the sticker shock in press coverage of the price hike is almost entirely about V4-Pro's reasoning-mode token consumption on harder tasks, not routine summarization on the cheaper model.
Where costs actually escalate: V4-Pro's default "thinking mode" on a complex question can burn several thousand reasoning tokens, billed at the output rate, before producing a short final answer. A single hard reasoning query at peak pricing ($3.96/M output tokens) generating 4,000 reasoning-plus-output tokens costs roughly $0.016, trivial on its own, but multiply that by a high-volume production application making thousands of such calls daily and it adds up fast. That's the scenario the price hike actually targets, and it's a completely different usage pattern from an individual chatting through the free app.
The self-hosting route, and its real cost
R1's open weights are free to download, but "free" only refers to the software license, not the hardware to run it. R1 is a large model; running it at usable speed requires either a multi-GPU workstation (realistically several thousand dollars of hardware even at the low end, using consumer cards with heavy quantization) or renting GPU time from a cloud provider, which reintroduces a metered cost, just paid to a different vendor than DeepSeek. Self-hosting makes sense for someone who already owns the hardware for other reasons, or who needs guaranteed data locality regardless of cost. It is not a free lunch for someone comparing it against the already-free chat app on convenience grounds.
How DeepSeek's free tier stacks up against ChatGPT, Gemini, and Claude
All four major chat assistants offer some free tier, but the shape of the free offering differs. ChatGPT's free tier runs on GPT-5-mini-class models with message-count rate limits that reset every few hours. Google's Gemini free tier (inside the Gemini app, separate from Gemini Notebook) similarly rate-limits by message count and gates its most capable model behind a paid plan. Claude's free tier from Anthropic works the same way, generous for casual use, throttled if you're a heavy daily user, with the strongest model reserved for paid tiers. DeepSeek's free tier stands out specifically because it gives free users its current flagship reasoning model, V4-Pro, not a scaled-down version, which is unusual among the four. The tradeoff, as covered above, is the data-residency and privacy profile that comes with a China-based company, not a capability downgrade.
Is DeepSeek free anywhere else?
OpenRouter, an API aggregator that's carried free-tier DeepSeek variants on and off since 2025, currently lists roughly 25 DeepSeek models (V4 Pro, V4 Flash, the full V3.x line, R1 and its distills), and as of this writing none are marked `:free`. That lineup rotates, so check at the time you're reading this rather than treating "no free option" as permanent. OpenRouter's general free tier, unrelated to DeepSeek specifically, caps out at 20 requests a minute and 50 a day, extendable to 1,000 a day with a one-time $10-plus credit purchase, but it doesn't currently apply to any DeepSeek model.
The privacy caveat, verified
DeepSeek's own privacy policy states it collects and stores personal data, every prompt, upload, plus device and network data, in the People's Republic of China, retained for as long as the account exists or longer for stated business interests. DeepSeek operates under China's 2017 National Intelligence Law, which can compel disclosure of user data to Chinese authorities regardless of where the user is physically located.
As of 2026, DeepSeek has been banned or restricted in Italy, South Korea, Taiwan, Australia, India, the Czech Republic, and at least 17 US states, with open regulatory investigations in France, Ireland, Germany, Belgium, and the Netherlands. A concrete, documented action: Berlin's data protection commissioner formally asked Apple and Google to remove DeepSeek from their app stores in Germany, stating the company hadn't provided convincing evidence that German users' data gets EU-equivalent protection once it reaches China. None of this determines whether DeepSeek is 'safe' for your specific use case, but it's the context that should inform the decision, especially for anything involving sensitive documents or client data.
One more thing worth flagging directly: deepseek.ai is not DeepSeek's real site, it's an independent, unaffiliated domain that ranks well in search and looks official. Trustpilot reviewers report it charging for things that are free on the real service. The only official DeepSeek domains are deepseek.com (corporate site), chat.deepseek.com (the free app), and platform.deepseek.com (the API dashboard). If a site asking for payment doesn't match one of those three, it isn't DeepSeek.
People also ask
Is the DeepSeek app completely free, or does it have a paid tier?
The chat app has no paid subscription tier at all. It's free on web, iOS, and Android, with no ads and no published hard daily message cap beyond fair-use throttling during high traffic.
What's the difference between DeepSeek's free chat and the DeepSeek API?
They're separate products with separate billing. The chat app at chat.deepseek.com is free with no subscription. The API at platform.deepseek.com is metered, pay-per-token, and used for building apps or integrations, not for casual chatting.
Did the August 2026 DeepSeek price increase make the free app more expensive too?
No. The price hike (up to 1,100% at peak hours, effective 16 August 2026) applies only to the developer-facing API. The free consumer chat app's access and limits are unchanged.
Can I use DeepSeek for free through OpenRouter or another platform?
Not currently. OpenRouter lists around 25 DeepSeek models but none are marked free as of this writing, though its free-model lineup has changed before and could again. Self-hosting R1's open weights from Hugging Face remains a genuinely free, if technical, alternative.
Is DeepSeek safe to use, given it's a China-based company?
DeepSeek's own privacy policy confirms data storage in China under a law that can compel disclosure to Chinese authorities. Multiple countries and US states have banned or restricted it as a result, and Berlin's data protection commissioner formally asked Apple and Google to remove it from their German app stores over unlawful data transfers. Weigh that against your specific use case, especially for sensitive data.
How many free tokens do new DeepSeek API accounts get, and do they expire?
Commonly reported as around 5 million tokens (roughly $8-9 of value), valid about 30 days, but this isn't documented as a guaranteed, permanent policy in DeepSeek's own docs, so check your account dashboard rather than assuming it applies.