Latest

AI assistants · reviewed 2026-09-16

DeepSeek logoDeepSeekdeepseek.com ↗

What it does

Free AI chatbot (web, iOS, Android) from Chinese lab DeepSeek, offering its flagship models, currently V4.1 Flash, at no cost, plus one of the cheapest developer APIs.

Strong fit

Pick it for a fully free, capable reasoning chatbot with no subscription, for a very cheap high-volume API, or when you need open weights you can run yourself.

Poor fit

Avoid for sensitive/confidential data (China-based service, trains on inputs by default) or if you need image/video generation and rich voice features.

Limitations

No paid consumer tier to buy priority or higher limits; inputs used for model training unless you opt out; service can throttle at peak demand; the API's cheapest rates apply off-peak only; self-hosting a 552B model needs serious hardware.

Privacy & data handling

Privacy policy: inputs used to train and improve the technology by default; opt-out via privacy@deepseek.com ().

Verified facts

  • free tier

    The chat product (chat.deepseek.com and the DeepSeek app) is free, including the newest flagship model, V4.1 Flash; no usage caps published.

    DeepSeek official site · checked 2026-09-16

  • pricing

    No paid chat plan, chat is free. Paid offering is the usage-based API. Published table, USD per 1M tokens: deepseek-flash $0.15-$0.30 input on a cache miss, $0.003-$0.006 on a cache hit, $0.60-$1.20 output; deepseek-v4-pro $0.66-$1.32, $0.022-$0.044 cache hit, $1.98-$3.96 output. Off-peak rates are half of peak; peak is 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday.

    DeepSeek API pricing · checked 2026-09-16

  • data handling

    Privacy policy: inputs used to train and improve the technology by default; opt-out via privacy@deepseek.com ().

    Official data-handling statement · checked 2026-07-29

  • open weights

    DeepSeek-V4.1-Flash weights are published on Hugging Face under the MIT licence: a 552B-parameter Causal Encoder-Decoder mixture-of-experts activating 8B parameters per token at prefill and 16B at decode, with a context window of up to one million tokens and image-and-text input. Deployment documented for vLLM, SGLang and Transformers.

    DeepSeek-V4.1-Flash model card · checked 2026-09-16

Wondering whether DeepSeek is the right pick for your task, or what to pair it with? Describe the outcome and get a full Blueprint.

Build my route

Facts above were checked on the dates shown and can change - confirm on the vendor’s own site before you rely on them.