Why it matters

Publishing competitive weights openly changes who gets to run a frontier-class model, not just who gets to call one.

DeepSeek

Added 2026-07-25

Open-weight Chinese models with strong reasoning and coding, available as chat and API.

DeepSeek interface

DeepSeek is a Chinese AI lab whose models compete on reasoning and coding while being published as open weights. The second half is the part that matters structurally: most labs at this capability level ship an API and nothing else, so anyone with a compliance constraint or a private-deployment requirement is locked out. Open weights remove that gate. In practice DeepSeek is worth reaching for on problems where you want the chain of reasoning visible, and worth knowing about as the fallback when a workload cannot leave your own infrastructure.