DeepSeek

Name: DeepSeek
Brand: DeepSeek

DeepSeek, China, chat.deepseek.com

A high-performance Chinese AI chatbot that is free to use. The models are also released as open source

Chatbot/Assistant Open Models

Go to official siteRequest a correction

CurrencyUSD KRW EUR

Personal starting priceFreeNo paid plan

Free tierFree

Korean supportGood

Price checked 2026-10-18, source: api-docs.deepseek.com/updates. Prices and benefits may change. Check the official site before you pay.

Pros

Chatbot fully free
Good at math and reasoning
API is very cheap

Cons

Controversy over overseas transfer of personal data
Slows down when traffic is heavy
Censorship of sensitive topics

These are opinions compiled by Pickx from public sources.

At a glance

DeepSeek is a chatbot and model service made by DeepSeek, an AI company in Hangzhou, China, and you can use it free on the web (chat.deepseek.com) and in the app. Since the DeepSeek-V4 series released in April 2026, all official services support a 1,000,000-token context by default, and the model weights are also published on Hugging Face. It suits people who want a high-performance chatbot for free and developers who need a very cheap API.

Full model list

Model	Type	Features	Where/plans available
DeepSeek-V4.1-Flash (API name `deepseek-flash`)	Text/reasoning/image understanding (multimodal)	Latest main model, released September 10, 2026 Upgraded to V4.1-Flash-1017 on October 17, 2026 (API name and price unchanged, web/app on the same version, shorter response time and output length) Native multimodal MoE model that understands images directly 1,000,000-token context Thinking mode Supports thinking effort (low/high/max)	Web/app free API pay-as-you-go Open weights (MIT, about 763B per Hugging Face)
DeepSeek-V4-Pro (`deepseek-v4-pro`, version V4-Pro-0813)	Text/reasoning/coding/agents	Top model with 1.6T total parameters (49B active) Strong at complex coding and agent work Max output 384K tokens API price cut in September 2026	Web/app (Expert mode and others) API Open weights (Hugging Face)
DeepSeek-V4-Flash V4-Flash-0731	Text/reasoning	Fast, cheap V4 model with 284B total (13B active) Replaced by V4.1-Flash from September Automatically routed to V4.1-Flash in the API	Previous model (API discontinued, weights published)
DeepSeek-V4-Flash-Vision-Exp	Image understanding (experimental)	Experimental vision model published in August 2026 Ended when merged into V4.1-Flash	Previous model
DeepSeek-V3.2 V3.2-Exp V3.2-Speciale	Text/reasoning	Main model in late 2025 Cuts long-context cost with sparse attention (DSA) Supports thinking and non-thinking modes in one model	Previous model (open weights, MIT)
DeepSeek-V3.1 V3.1-Terminus	Text/reasoning	First hybrid model combining thinking and regular modes Released August to September 2025	Previous model (open weights)
DeepSeek-R1 R1-0528	Reasoning	Reasoning model (671B MoE) that became a worldwide sensation in January 2025 Released together with small distilled models (1.5B to 70B)	Previous model (open weights, MIT)
DeepSeek-V3 V3-0324	Text	General-purpose MoE model with 671B (37B active) The model that started the low-cost, high-performance trend	Previous model (open weights, MIT)
DeepSeek-OCR OCR-2	Document recognition (OCR)	Research model that reads document images as compressed visual tokens OCR is 3B	Open weights (Hugging Face)
DeepSeekMath-V2 DeepSeek-Prover-V2	Math/theorem proving	Math proof verification Formal theorem proving (Lean) Research models specialized in these two areas	Open weights
Janus-Pro	Image understanding/generation	Small multimodal model that handles image understanding and generation together Released January 2025	Open weights (not in the chat service)

Note: In the web and app you do not pick a model directly, but choose Expert mode / Instant mode. Which model each mode connects to is not disclosed in detail officially (needs checking).

Summary by plan

Plan	Price	What you can do	Usage limits	Features
Free (web/app)	$0	Chat Thinking mode (Expert) Web search File and image upload	Not published officially May slow down or be limited when many users are on	No paid personal plan
API – deepseek-flash (V4.1-Flash)	Per 1,000,000 tokens: input $0.15 (peak $0.30) Cache-hit input $0.003 (peak $0.006) Output $0.60 (peak $1.20)	Image input Tool calls OpenAI/Anthropic-compatible API Responses API Web search Search Fetch API (released 2026-10-17)	1,000,000-token context Usage limit not published officially (pay-as-you-go)	Among the lowest prices in the industry Prepaid top-up
API – deepseek-v4-pro	Per 1,000,000 tokens: input $0.66 (peak $1.32) Cache hit $0.022 (peak $0.044) Output $1.98 (peak $3.96)	Choose thinking effort	1,000,000-token context Max output 384K	For hard coding and agent work
Peak/off-peak pricing	Off-peak is half of peak	Peak: weekdays 01:00-04:00, 06:00-10:00 UTC (Korea time 10:00-13:00, 15:00-19:00) Weekends and Chinese public holidays are all off-peak	Not applicable	Applies from August 16, 2026
API – Search / Fetch	$0.75 per 1,000 successful requests (billing starts 2026-11-01 at 0:00 Beijing time)	Get web search results Fetch web page content	5 requests per second each for Search and Fetch per account Fetch has additional per-domain limits	Failures, empty results and timeouts are free Server-side calls via DeepSeek Harness, Claude Code and Codex are currently free
For business	No separate plan	Use the API as is Install the published weights on your own server or cloud	Not published officially	If data security matters, self-hosting (on-premises) is an alternative

Which plan for whom

Free (web/app): People who want to try math, coding and long-document summaries without paying. However, the servers are in China, so we recommend not entering personal information or company secrets.
API – deepseek-flash: Developers who want to run chatbots, summaries, classification and other bulk processing cheaply. Running it in off-peak hours (evening to dawn in Korea time, and weekends) halves the cost.
API – deepseek-v4-pro: Teams that need coding agents or complex reasoning but find the cost of top US models a burden.
Self-hosting (open weights): Companies and institutions that cannot send data outside. The MIT license allows free commercial use, but V4-class models need large GPU servers.

Notes

Information checked: 2026-10-18
Sources: DeepSeek API pricing, DeepSeek API change log, Search / Fetch API guide, DeepSeek-V4 announcement, Hugging Face deepseek-ai, DeepSeek chat

Good for

2 views

Prices converted at 1 USD = ₩1,329.49 (2026-10-19). Original price shown in brackets. Taxes as listed by each provider.

Comments

No comments yet. Be the first.

Your comments and nickname are public. If you see abuse or advertising, report it here and we will review and remove it.