Free DeepSeek API in 2026: What Is Actually Free

DeepSeek's own API has no free tier: you top up a balance and pay per token, and off-peak hours cost half the peak price. That is a discount, not free. To call a DeepSeek model for $0 you go through a provider that hosts the open weights and gives them away within a quota. Every number below was checked on DeepSeek's docs or the provider's own pages on 2026-10-21.

DeepSeek's own API: cheap, not free

DeepSeek's pricing page lists two model names: deepseek-flash (DeepSeek-V4.1-Flash-1017) and deepseek-v4-pro (DeepSeek-V4-Pro-0813), both with a 1M-token context and up to 384K output. Prices per 1M tokens, off-peak / peak:

• Peak hours: "01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday, excluding Chinese public holidays"; weekends and Chinese public holidays are off-peak all day, and "Off-peak rates are half of the peak rates". The DeepSeek peak hours tool shows those windows in your time zone.
• No sign-up bonus is stated. The billing rules mention a "granted balance" without saying who gets one or how much. With an empty balance the API returns 402 Insufficient Balance, so plan on a top-up. Top-ups do not expire and unused balance can be refunded (FAQ).
• Rate limits are concurrency caps: 2,500 concurrent requests per account for deepseek-flash and 500 for deepseek-v4-pro, HTTP 429 above that; raising them is free on request (rate limits).
• Old model names: DeepSeek announced that deepseek-chat and deepseek-reasoner would be discontinued on 2026-07-24 (updates); current docs use deepseek-flash. The later names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, but they are served by DeepSeek-V4.1-Flash-1017 and billed at the Flash price.
• The chat app at chat.deepseek.com is a separate product from the API; using it does not give you API access.

Where DeepSeek models are free

Two more give a small credit you can spend on DeepSeek, not a free model: SiliconFlow gives $1 to start (DeepSeek-V4.1-Flash is $0.15 in / $0.60 out per 1M tokens there), and Nebius Token Factory gives $1 valid for 30 days and needs a bank card for billing.

Which one to pick

• Most requests a day: NVIDIA NIM (10,000 a day) or Infron (1,000 a day), both on V4.1 Flash with the long context.
• No balance, no card: Routeway or Token Harbor. Routeway's free variant has a 42K context, so long documents need another provider.
• A fixed token budget: LLM7 at 100,000 tokens a day, or Scaleway's one-time 1M tokens hosted in Europe.
• Production traffic: DeepSeek's own API. Run batch jobs off-peak to pay half; the peak hours tool shows when.

Steps

NVIDIA NIM

1. Sign in at build.nvidia.com and create an API key.
2. Call the OpenAI-compatible endpoint: curl https://integrate.api.nvidia.com/v1/chat/completions -H "Authorization: Bearer $NVIDIA_API_KEY" -H "Content-Type: application/json" -d '{"model": "deepseek-ai/deepseek-v4.1-flash", "messages": [{"role": "user", "content": "Hello"}]}'

Infron or Routeway

1. Create an account and an API key (Infron also needs a $5 balance on the account).
2. Point any OpenAI SDK at https://llm.onerouter.pro/v1 (Infron) or https://api.routeway.ai/v1 (Routeway) and use the :free model ID from the table.

DeepSeek's own API

1. Sign up at platform.deepseek.com, top up, and create a key on the API Keys page.
2. Use base URL https://api.deepseek.com with the OpenAI SDK, or https://api.deepseek.com/anthropic with the Anthropic SDK, and model deepseek-flash.

The code is the same on every provider; only the base URL, key and model ID change. That makes it easy to start on a free quota and move to DeepSeek's paid API later.

Self-hosting

Both current models are open weights under the MIT License: DeepSeek-V4.1-Flash-1017 and DeepSeek-V4-Pro-0813. The Flash weights alone total about 763B parameters, so running them yourself needs a multi-GPU server; for most people a hosted free quota is the practical route.

Limits and the catch

• Off-peak is a discount, not free. DeepSeek charges every token; peak hours cost twice the off-peak rate.
• Free variants are not always the same model. Routeway's free DeepSeek has a 42K context, and LLM7 only gives the older V4 Flash 0731 for free. Check the model ID before you rely on long context.
• Free lineups rotate. Infron, Routeway, Token Harbor and AMD all say free models can change or be withdrawn; keep the base URL and model ID in config, not code.
• Read the data terms. Token Harbor may keep free-model prompts for product improvement; AMD logs prompts and outputs (two days) but says it does not train on them.
• Other free LLM APIs: for Groq, Gemini or OpenAI models see the free Groq API key, free Gemini API key and free OpenAI API access guides, or compare them all on free LLM APIs.