DeepSeek: Is It Free, What V4.1 Flash Is and What the API Costs

DeepSeek is a Chinese AI assistant you can use on the web at chat.deepseek.com and in its iPhone and Android apps. Is DeepSeek free? Yes: the chat costs nothing, you only need an account, and there's no paid plan on its front page or its App Store listing. What you do pay for is the API, if you're a developer: it's pay-as-you-go in US dollars, and its V4.1 Flash model costs $1.20 per million output tokens at peak hours and half that off-peak.

What DeepSeek is in 2026

For most people, it's a chatbot that competes with ChatGPT and doesn't charge you for it. For developers, it's a family of models you can call through an API or download and run yourself, because DeepSeek publishes the weights.

The front page makes the split obvious. Under a prompt box that says "Ask anything, explore together," with DeepThink and Search buttons, there are three ways in: Chat with DeepSeek, API Platform and Harness Desktop. The Get App link up top leads to the mobile apps.

The model family has moved fast. After V4 Flash and V4 Pro came V4.1 Flash, and that's the one most people will touch. On October 17, DeepSeek released an update called V4.1-Flash-1017, and its front page says it brings "significantly enhanced project development capabilities, enabling tasks to be completed at lower cost and faster speed."

Is DeepSeek free?

The chat, yes. On the US App Store, "DeepSeek - AI Assistant" is listed as Free, rated 13+, and its own description promises the official assistant "for free." Neither the front page nor the App Store listing shows a subscription or an upgrade.

There are two small hurdles. First, you need an account, with your email, phone, Google or Apple. Second, once you're inside, DeepSeek asks for your year and month of birth before it lets you send a single message, for age verification. When I opened chat.deepseek.com from my test browser, it didn't even get that far: it first asked me to "confirm you are human" with a security check.

V4.1 Flash, DeepSeek's current model

Sorry, this part gets a little technical. DeepSeek's API offers two models today. The everyday one is called deepseek-flash, and it currently runs DeepSeek-V4.1-Flash-1017. The bigger one is deepseek-v4-pro. Both handle a context of 1 million tokens, both can "think" before answering (that's the default, and you can switch it off), and both work with the OpenAI and Anthropic API formats, so most tools that already talk to those can talk to DeepSeek by changing a URL. Flash can read images; Pro can't.

If you still have code calling deepseek-v4-flash, it keeps working: DeepSeek says the models behind those old names are retired, and their calls are now served by V4.1-Flash-1017 at the Flash price.

The bigger news for developers is that the weights are public. On Hugging Face, DeepSeek-V4.1-Flash-1017 is listed under the MIT license, one of the most permissive there is, and DeepSeek describes it as "a minor update to DeepSeek-V4.1-Flash" with gains in reasoning, coding, agentic tasks and image understanding, plus shorter outputs.

A reality check: this model is huge. Open weights don't mean it runs on your laptop. For that, my older guide on running DeepSeek R1 locally covers the smaller versions. What open weights do give you is choice: other companies host V4.1 Flash too. On OpenRouter, the 1017 version shows up with the same 1 million token context, a release date of October 17, 2026, and a list of providers that serve it at their own prices.

DeepSeek API pricing

These are the prices on DeepSeek's API docs, in US dollars per million tokens. "Cache hit" is input DeepSeek has already seen, like a long system prompt you send every time, and it's billed at a fraction of the normal rate.

Off-peak is exactly half of peak, and here's the quirk: peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday. On the US East Coast in daylight time, that's 9 pm to midnight and 2 am to 6 am, from Sunday night to Friday morning (an hour earlier once clocks fall back on November 1, and Chinese public holidays are off-peak all day). So if you work US office hours, your calls already land at the cheaper rate, and so does almost all of the weekend.

There's no subscription. You top up a balance and DeepSeek deducts each call from it, and any promotional credit it grants you gets spent first.

The newest addition is Search and Fetch, two APIs that let the model search the web and read pages. Each costs $0.75 per 1,000 successful requests. If you use DeepSeek through its Harness app, Claude Code or Codex, those searches are free for now, but DeepSeek says per-request billing starts on November 1, 2026, Beijing time.

What I could see without an account

Most of it, honestly. The front page, the US App Store listing, the full API pricing page and the model pages on Hugging Face and OpenRouter are all public. The one closed door is the chat itself: my test browser got a security check, and once you're signed in, DeepSeek asks for your birth date before your first message. How DeepSeek answers is something you can see for yourself in a couple of minutes once you sign up.

Who DeepSeek is for, and who should skip it

DeepSeek makes sense if you want a chatbot without a monthly bill, or if you're a developer watching costs: a million output tokens on Flash costs less than a coffee, and the OpenAI-compatible format makes it easy to try in an existing project. The open weights are a real plus if your company needs to host its own model.

Skip the chat, or at least be careful with it, if your conversations include anything private, since everything is stored in China. And if your goal is to run a model on your own computer, V4.1 Flash is the wrong size; look at the smaller models instead.

DeepSeek or something else?

ChatGPT · OpenAI

Free with limits · from $8/mo

ChatGPT review →

If you want the most polished all-rounder, ChatGPT is free too, lets you start without an account, and Plus costs $20 a month. Claude is my pick for long writing and code, with a free plan and Pro at $20 a month including Claude Code. And if your life runs on Gmail and Google Docs, Gemini is free with a Google Account, and Google AI Pro is $19.99 a month.