Is DeepSeek Free? What You Actually Pay in 2026

Short answer: yes, the DeepSeek chatbot is free to use in 2026, and the API charges per token with no fixed subscription. That’s the clean version. The messy version — the one you need before you commit an account, a workflow, or a budget — has layers: the web chat, the mobile app, the paid API with its peak and off-peak rates, the open weights you can self-host, and the regulatory quirks that affect whether “deepseek free” even means “available to you” where you live. This guide walks through each layer with prices as of October 2026, what’s included, and where the hidden costs sit. By the end, you’ll know which surface is genuinely free, which looks free but isn’t, and how to pick the right one.

The short answer: which parts of DeepSeek are free

DeepSeek operates on three distinct surfaces, plus the open weights, and two of them are free at the point of use:

• Web chat at chat.deepseek.com — free, no subscription.
• Mobile app (iOS and Android) — free; on 24 October 2026 neither store listing showed in-app purchases.
• Developer API — paid, per token, at peak and off-peak rates. No monthly fee, but no free tier either beyond any granted balance the billing console may show.
• Self-hosted open weights — the weights are free to download; you pay for the hardware to run them.

If you just want to chat with a capable model without a credit card, the web and app are the answer. If you want to build software on top of DeepSeek, you pay per million tokens. The rest of this article unpacks each case with current numbers.

The free chatbot: what you get at chat.deepseek.com

According to DeepSeek’s 17 October 2026 Change Log, the official web and app models were updated to DeepSeek-V4.1-Flash-1017, the model behind the API’s deepseek-flash; its 13 August 2026 release notes said V4 Pro is available on the app and web via “Expert Mode”. DeepSeek’s homepage, checked on 24 October 2026, pictures a chat box with DeepThink and Search buttons, but DeepSeek does not document the chat’s modes in detail, so check the interface for what is offered today.

There is no subscription paywall on the chat. DeepSeek does not publicly document a daily message cap on the web chat as of October 2026; its Rate Limit page covers only API concurrency. That does not mean infinite throughput — during traffic surges the service can slow down or queue requests, and DeepSeek publishes incidents on its status page.

If you want an account, see our guide to signing up for DeepSeek for the current steps.

The mobile app: also free, with caveats

The official DeepSeek app is free on both stores. On 24 October 2026 the App Store listing (“DeepSeek – AI Assistant”, seller Hangzhou DeepSeek Artificial Intelligence Co., Ltd, iOS 15.0 or later) showed no in-app purchases, and the Google Play listing (developer “DeepSeek”) showed neither in-app purchases nor a “Contains ads” label. According to the 17 October 2026 Change Log, the app runs the same updated model as the web. Before you install, check two things:

1. Availability. Italy’s Garante restricted the DeepSeek app on privacy grounds in January 2025, and several US states restrict it on government devices. See our DeepSeek availability by country reference for the current picture in your jurisdiction.
2. Authenticity. The App Store and Play Store host imitator apps that charge subscriptions for a wrapper around the free service. Our guide to verifying the official DeepSeek app shows you how to check the seller, the price and the download link before installing.

The API: not free, but inexpensive per token

This is where the word “free” gets slippery. The DeepSeek API charges per million tokens and has no flat subscription. DeepSeek may offer a granted balance — a small promotional credit that can expire — and its pricing page says fees are deducted from the granted balance first when both a granted and a topped-up balance exist. There is no published, guaranteed free tier, so third-party claims of a fixed free token allowance do not reflect the official documentation. Check your billing console for current offers.

Chat requests hit POST /chat/completions, the OpenAI-compatible endpoint at https://api.deepseek.com. DeepSeek also supports the Anthropic API format at a separate base URL, https://api.deepseek.com/anthropic, so either SDK works by setting the matching base_url and api_key. The API is stateless — unlike the web chat, which keeps your session on DeepSeek’s side, every API call must resend the full conversation history. For request details and code samples, see the DeepSeek API documentation guide.

Pricing snapshot (October 2026)

According to DeepSeek’s pricing page as of October 2026, in USD per 1M tokens:

Off-peak rates are exactly half the peak rates ($0.30 ÷ 2 = $0.15; $1.20 ÷ 2 = $0.60; $1.32 ÷ 2 = $0.66; $3.96 ÷ 2 = $1.98). Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays; every other hour, including weekends and Chinese public holidays in full, is off-peak. The 10 September 2026 V4.1-Flash release reduced API prices (effective 04:00 UTC that day), and the 17 October 2026 update left them unchanged; the pricing page notes that prices may vary.

deepseek-v4-pro costs 3.3× deepseek-flash per output token ($3.96 ÷ $1.20 = 3.3), 4.4× on cache-miss input ($1.32 ÷ $0.30) and about 7.3× on cache-hit input ($0.044 ÷ $0.006), in both periods, and only Flash accepts images — so default to Flash unless you have measured a quality gap that matters.

Two points the pricing table doesn’t shout about:

• Cache-hit pricing is automatic: repeated prefixes in your messages are billed at the cache-hit rate without an opt-in, but the uncached part of each request is still billed at the cache-miss rate.
• Off-peak pricing is back. DeepSeek ended its earlier off-peak discounts at 16:00 UTC on 5 September 2025 and introduced the current peak/off-peak rates, off-peak at half price, from 16:00 UTC on 16 August 2026. Scheduling batch work outside peak hours halves its cost.

Legacy model IDs and the retirement date

If you have older code using deepseek-chat or deepseek-reasoner, check it now: both IDs were scheduled for retirement on 24 July 2026 at 15:59 UTC and are no longer listed in the official docs (until then they pointed to deepseek-v4-flash in non-thinking and thinking mode). The legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, served by DeepSeek-V4.1-Flash-1017 and billed at the deepseek-flash price. Migration is a one-line model= change to deepseek-flash or deepseek-v4-pro; the base_url stays the same. See our DeepSeek API pricing page for the full migration note.

A worked example: what 1 million calls actually cost

Imagine a chatbot on deepseek-flash with a 2,000-token system prompt (cached across calls), a 200-token user message (uncached on each call), and a 300-token reply, run 1,000,000 times. That is 2,000,000,000 cached input tokens, 200,000,000 uncached input tokens and 300,000,000 output tokens:

deepseek-flash          PEAK                         OFF-PEAK
Input, cache hit  : 2,000M × $0.006/M = $12.00    2,000M × $0.003/M = $6.00
Input, cache miss :   200M × $0.30/M  = $60.00      200M × $0.15/M  = $30.00
Output            :   300M × $1.20/M  = $360.00     300M × $0.60/M  = $180.00
                                        -------                       -------
Total                                   $432.00                       $216.00

On deepseek-v4-pro the same workload costs $1,540.00 if run entirely in peak hours ($88.00 cached input + $264.00 uncached input + $1,188.00 output) or $770.00 entirely off-peak ($44.00 + $132.00 + $594.00) — about 3.6× the Flash bill in either period ($1,540.00 ÷ $432.00). A real workload usually straddles both periods: if, say, half the tokens fall in each, the Flash total is $324.00 ($216.00 + $108.00). If you’re running DeepSeek for business workloads at volume, that ratio drives the decision. The DeepSeek cost estimator works out your own mix.

The open weights: free to download, not free to run

Both current models publish their weights under the MIT license on Hugging Face: DeepSeek-V4.1-Flash-1017 and DeepSeek-V4-Pro-0813. You can download either and run it yourself — no licence fee.

The catch is hardware. The current files total about 510 GB for DeepSeek-V4.1-Flash-1017 and about 893 GB for DeepSeek-V4-Pro-0813, and the DeepSeek-V4-Pro-0813 model card’s vLLM example serves it on a 4×GB300 GPU node. In practice, serious use of the full-size weights means a multi-GPU server or a rented cloud instance. Our install DeepSeek locally walkthrough and DeepSeek hardware calculator go deeper on the real cost of self-hosting.

Licensing is per-model, so verify before you ship a product:

Free vs paid: which layer fits which user

If you’re still weighing these against each other, our dedicated DeepSeek free vs paid comparison lays out the trade-offs in more detail.

Hidden costs of “free”

“Free” is never only about money. Three things to weigh:

Privacy

Chats submitted to DeepSeek’s web and app are processed on servers in China and may be stored under Chinese law. That includes any file you upload and any code you paste in. For non-sensitive work this is a reasonable trade; for client data, legal documents, or proprietary code, it is not. See DeepSeek privacy for the full breakdown, and consider the self-hosted route if data residency matters.

Availability

Italy restricted the app in January 2025 over data-protection concerns. Several US states restrict its use on government devices, and a US Department of Homeland Security order published on 20 October 2026 excludes DeepSeek products from federal civilian agency IT; none of these measures bans private use. Check our DeepSeek US restrictions page before assuming you can deploy it in a regulated context.

Reliability

The free web chat goes through traffic surges, especially around new releases, and the API enforces per-account concurrency limits (2,500 concurrent requests for deepseek-flash, 500 for deepseek-v4-pro). For anything business-critical, watch the status page and budget for a fallback provider.

So, is DeepSeek actually free?

For end users: yes. The web chat and the mobile app are free, with no subscription. You pay with data — your prompts are processed in China — and with the occasional slowdown during busy periods.

For developers: not really. The API is inexpensive — $0.30 / $1.20 per 1M cache-miss input / output tokens at peak on deepseek-flash, half that off-peak — but it is priced per token. Promotional credits should be confirmed in your billing console, not assumed from third-party blog posts.

For researchers and companies that need data residency: the open weights under MIT are the closest thing to a truly free path, at the cost of the hardware to run them.

Whichever surface fits you, start with the DeepSeek beginner guides and work up. The chat is fine to pick up in five minutes; the API rewards spending an hour on the docs before you burn tokens. For JSON output, the JSON mode guide covers the edge cases.