deepseek-r1.com is an independent, third-party information and demo hub built around DeepSeek's R1 reasoning model. It is not operated by DeepSeek, but it collects the material people usually have to hunt for across many pages: the architecture explanation, benchmark numbers, distilled model downloads and a rate-card comparison against a well-known closed reasoning model.,The most practical feature is the free web chat, which needs no account and puts a reasoning assistant in front of visitors within a couple of clicks. A second, technically interesting path runs a small distilled checkpoint (DeepSeek-R1-Distill-Qwen-1.5B) locally through WebGPU, Transformers.js and ONNX Runtime Web, so prompts never leave the browser and can even keep working offline once loaded.,For builders, the site documents the Mixture-of-Experts layout with roughly 37B active and 671B total parameters, 128K context support, an OpenAI-compatible API endpoint and the MIT-licensed distilled family from 1.5B up to 70B. That combination makes it a convenient starting point for evaluating whether R1 fits a project before committing to a self-hosted or hosted deployment.