LIVE
EU AI Act enforcement begins · June 2026NIST AI RMF — risk management framework publishedISO/IEC 42001 AI management standard now certifiableOpenAI o3 sets new reasoning benchmarksAnthropic raises $4B Series EEU AI Act enforcement begins · June 2026NIST AI RMF — risk management framework publishedISO/IEC 42001 AI management standard now certifiableOpenAI o3 sets new reasoning benchmarksAnthropic raises $4B Series EEU AI Act enforcement begins · June 2026NIST AI RMF — risk management framework publishedISO/IEC 42001 AI management standard now certifiableOpenAI o3 sets new reasoning benchmarksAnthropic raises $4B Series E

DeepSeek

Freemium
Featured
API Available

China's open-source LLM that shocked the AI industry on cost and performance

DeepSeek AI · Language Models

Launched Jan 20, 2025 (R1); Dec 26, 2024 (V3)

Visit Website

Quick Answer

DeepSeek is china's open-source llm that shocked the ai industry on cost and performance, made by DeepSeek AI. It offers a free tier with paid upgrades. Key uses: Code generation, Mathematical reasoning, Research.

Last updated:

Users

~20M+ registered users (Feb 2026 estimate); #1 App Store globally Jan 2025

Valuation

Reported $5B+ valuation; backed by High-Flyer Capital

What is DeepSeek?

DeepSeek R1 and V3 are open-source large language models from Chinese lab DeepSeek AI that match or exceed GPT-4-class performance at a fraction of the training cost. DeepSeek R1 is a reasoning model trained with reinforcement learning; V3 is a 671B mixture-of-experts model. Both are MIT-licensed and freely downloadable.

How did DeepSeek get started?

Founded 2023 by Liang Wenfeng, a quantitative hedge fund billionaire. DeepSeek-V2 (May 2024) shocked with MoE efficiency. DeepSeek-V3 (Dec 26, 2024) trained on 2.8M H800 GPU-hours at ~$5.6M cost. DeepSeek-R1 (Jan 20, 2025) matched OpenAI o1 on benchmarks. App went viral globally, caused major stock market movements. Sparked US Congressional hearings on AI competitiveness.

What's new in DeepSeek in 2026?

DeepSeek R2 reported in development (Jun 2026); DeepSeek V3-0324 update (Mar 2026) improves coding; DeepSeek Prover V2 (May 2025) formal math; App reached #1 in App Store globally Jan 2025; Caused 'AI scare' in US market, Nvidia stock -17% in one day

How much does DeepSeek cost?

Chat App

$0

  • DeepSeek V3 / R1 access
  • Web + mobile app
  • Deep thinking mode
  • Web search

API — Cache Hit

$0.07/M input tokens

  • V3 & R1 API
  • Cache hit pricing
  • OpenAI-compatible endpoint

API — Input

$0.27/M tokens (V3) / $0.55/M (R1)

  • Full V3/R1 model
  • Reasoning traces
  • Function calling

API — Output

$1.10/M tokens (V3) / $2.19/M (R1)

  • Output token pricing
  • Reasoning output included

What can you do with DeepSeek?

  • Code generation
  • Mathematical reasoning
  • Research
  • Cost-efficient inference
  • Self-hosting

Pros

  • MIT-licensed open weights
  • Matches GPT-4o performance
  • Fraction of training cost ($5.6M)
  • Excellent coding & math
  • Self-hostable

Cons

  • Chinese jurisdiction/data concerns
  • Chat UI less polished
  • Censorship on sensitive topics
  • Self-hosting requires significant hardware

Frequently Asked Questions about DeepSeek

How much does DeepSeek cost?

DeepSeek pricing: Chat App at $0; API — Cache Hit at $0.07/M input tokens; API — Input at $0.27/M tokens (V3) / $0.55/M (R1); API — Output at $1.10/M tokens (V3) / $2.19/M (R1).

Is DeepSeek free?

DeepSeek offers a free tier with limited usage. Paid plans unlock additional features.

What is DeepSeek used for?

DeepSeek is used for: Code generation, Mathematical reasoning, Research, Cost-efficient inference, Self-hosting.

What are the best alternatives to DeepSeek?

Top alternatives to DeepSeek include ChatGPT, Claude, Mistral AI, Llama.

What platforms does DeepSeek support?

DeepSeek is available on: Web, API, Self-hosted, Hugging Face, Ollama.

Details

Vendor

DeepSeek AI

Category

Language Models

Pricing Model

Freemium

Launched

Jan 20, 2025 (R1); Dec 26, 2024 (V3)

Underlying Models

DeepSeek R1 (671B MoE)DeepSeek V3 (671B MoE)DeepSeek R1-Distill-Llama-70BDeepSeek V2.5

Platforms

WebAPISelf-hostedHugging FaceOllama

Integrations

OllamavLLMLangChainLlamaIndexOpenRouterAWS BedrockAzure AI

Ideal For

  • Researchers
  • Cost-conscious enterprises
  • Self-hosted deployments
  • Mathematical reasoning
  • Coding tasks

Tags

Open-SourceReasoningChinese AIEfficientMoEMIT License