FLASHAlibaba starts a major restructuring: e-commerce merged, cloud chip unit integrated, Qwen spun off as a separate subsidiary1h ago
← DIRECTORY
Foundation Models

DeepSeek

Open-weight frontier models built at a fraction of the usual cost

Founded 2023Hangzhou, China🔓 Open-sourceWebsite ↗

DeepSeek is a Hangzhou-based AI lab incubated by the quantitative hedge fund High-Flyer. Its open-weight models rival top proprietary systems while training at far lower cost. The V4 generation, released in 2026 under an MIT license, is among the largest open-weight models available and is text-only.

Models

  • DeepSeek-V4-ProFlagship open-weight MoE model (~1.6T params), 1M-token context💬 Text🧮 Reasoning💻 Code
  • DeepSeek-V4-FlashFaster, lighter V4 variant for high-volume use💬 Text🧮 Reasoning💻 Code

Highlights

  • V4 (2026) is among the largest open-weight models, released under a permissive MIT license
  • Frontier-level performance at a fraction of typical training cost; widely used by developers worldwide

Information compiled from public sources; refer to the company's official website for authoritative details.