What DeepSeek does
DeepSeek is a frontier open-weight AI lab developing large language and reasoning models, plus an ecosystem of deployment products for consumers (web/app), developers (API), and local/agent builders (DeepSeek Harness). The company positions its model releases as open-sourced / open-weight via a Transparency Center, and it emphasizes long-context efficiency and agent-oriented tool use across its V4 and V3 model series.
More
DeepSeek’s current flagship family is the V4 series, with DeepSeek-V4-Pro and DeepSeek-V4-Flash marketed as million-token-context models and served through its web/app experiences and through the DeepSeek API platform. DeepSeek also publishes technical reports and “model cards” and provides model/version documentation through its Transparency Center and research/news pages. DeepSeek’s business model is primarily usage-based: it sells access to its models through the DeepSeek API (token-based pricing, including cache-hit vs cache-miss distinctions) and supports self-host/open-weight usage for users who choose that route.