DeepSeek
Chinese AI lab's open-weight models with frontier reasoning at ultra-low cost
Free and open-weight, API input ~$0.14/M tokens, output ~$0.28/M tokens
Overview
DeepSeek is a Chinese AI lab producing open-weight frontier models that compete with GPT-4 and Claude at a fraction of the cost. DeepSeek-V3 and R1 stunned the AI community with their performance on coding and reasoning benchmarks, achieved with significantly less compute than Western counterparts.
Key Features
- DeepSeek-R1: reasoning model competitive with OpenAI o1 on math and coding
- DeepSeek-V3: general-purpose model with 671B MoE parameters
- Fully open weights — downloadable and self-hostable
- API access at a fraction of comparable US model prices
- Strong multilingual support, especially Chinese and English
- MIT license for commercial use
Pricing: Free web chat at chat.deepseek.com; API at ~$0.14/1M input tokens for V3.
Pros
- Ultra-low API pricing significantly undercuts all competitors
- R1 reasoning model competitive with o1 on major benchmarks
- Fully open-weight with permissive licensing
- Strong at math and code with dedicated specialist models
Cons
- Primarily optimized for English and Chinese
- Subject to Chinese regulatory requirements
- Smaller ecosystem and fewer integrations than Western competitors
Tags
Product Updates
Similar Tools
Qwen
Alibaba's comprehensive open-weight family spanning text, code, vision, audio, and speech
GLM (Zhipu AI)
Chinese AI lab with open-weight models spanning text, vision, voice, and video generation
Llama
Meta's open-weight family spanning Llama 3.1 (8B–405B) and Llama 3.2 (1B–90B vision)
Mistral
French AI lab offering frontier models from edge to enterprise with open-weight options
