Independent hosted chat and API service

Use DeepSeek for chat or API workflows

Ask a question in the browser, turn on web search when you need current sources, or attach a document whose text you want to analyze. Developers can use three familiar API formats.

DeepSeek-V4.io is an independent service, not the official DeepSeek website. An account is required. API keys require a paid one-time usage pack, and messages or extracted text may be processed by third-party providers.

Chat preview
Live example
What would you like to work on today?
Compare two approaches and cite current sources.
Assistant is typing…
Describe your task and continue in chat…
What you can do

One workspace for chat, current sources, files, and API access

Use the same account balance across hosted chat and API workflows. Turn on search or attach a document only when the task needs it.

Hosted chat

Ask for explanations, summaries, writing, or help with code in a saved conversation.

Multi-turn conversations
Markdown downloads
Language preference
10K
signup balance
3
API formats
5
files per upload
5 MB
per file

Getting started

Start with a short, real task

  1. 1

    Create your account

    Sign in with Google or verify your email. A new account receives a 10K signup balance for trying the hosted chat.

  2. 2

    Enable only what you need

    Use Web Search for current information and attach files when document text is part of the question. Neither is required for a normal conversation.

  3. 3

    Check the answer and sources

    Open primary sources before making an important decision. If search fails, the interface identifies that the response was not verified on the web.

Connect applications and coding agents

After purchasing a one-time usage pack, create an API key in the Developer Dashboard. The base URL is https://deepseek-v4.io/api/v1 and the public model ID is deepseek-v4-flash. A short, billed connection test is available after key creation.

Chat Completions

OpenAI-compatible chat, streaming, and tool calls.

Responses

Responses API format with tool-call support.

Anthropic Messages

Anthropic Messages request format without claiming complete equivalence with the official service.

Important limits

Know the boundaries before you use it

  • DeepSeek-V4.io accounts, API keys, balances, and billing are separate from official DeepSeek and OpenRouter accounts.
  • Chat and API consume this site’s balance according to actual processing. Long conversations and attached documents can increase usage.
  • Conversations are saved to the account. Inputs and extracted text are processed by the configured model provider; search providers process the query when Web Search is enabled. Do not submit confidential information.
  • Each file can be up to 5 MB, with a 10 MB combined limit and up to five files. Very long extracted text may be truncated.
  • Image input, image generation, and OCR for scanned PDFs are not currently available.
  • Model answers and search results can be wrong. Verify primary sources and seek qualified advice for legal, medical, or financial decisions.

Frequently asked questions

Clear answers about access, billing, files, and data handling.

Still have questions? Contact support

New accounts receive a 10K signup balance for trying the hosted chat. Creating an API key requires a paid, one-time usage pack.

No. DeepSeek-V4.io is an independent gateway. Its account, API keys, and balance are separate from official DeepSeek and OpenRouter accounts.

Only when a PDF contains a readable text layer. The current upload flow does not run OCR on image-only or scanned pages.

Enable Web Search when a question needs current information. The response shows the sources returned by the search provider. Always open primary sources before making an important decision.

Paid usage packs add a one-time balance that can fund hosted chat and API requests. There is no recurring subscription. Actual usage depends on the amount of text processed.

Conversations are saved to your account. Messages and extracted document text are processed by the configured model provider, and search queries are processed by the search provider when Web Search is enabled. Do not submit secrets or sensitive personal data.

Try one real question

Start with a short task and evaluate the answer, response time, and—when search is enabled—the cited sources. If you need an integration, create a key after purchase and run the connection test before changing your application.