DeepSeek is the Chinese AI lab whose R1 release in early 2025 reset expectations for what open models cost. The current generation pairs DeepSeek-V4-Pro for frontier reasoning with DeepSeek-V4.1-Flash (vision-capable), both with 1M-token context windows and off-peak API pricing at roughly half of peak rates.
The chat app at chat.deepseek.com stays free with full model access, thinking mode and web search. Developers get an OpenAI-compatible API (plus an Anthropic-format endpoint), 25-hour off-peak windows and concurrency up to 2,500 – the reason DeepSeek remains the default price-performance benchmark every other lab is compared against.