DeepSeek

A Chinese AI lab and llm-provider notable for disrupting the bottom of the market: ultra-low llm-api-pricing behind OpenAI- and Anthropic-compatible APIs, plus strong open-weight-models. The seed subject of this wiki (its deepseek-api-docs was the founding source).

Models (2026)

• DeepSeek V4 Pro / V4 Flash — 1M-token context (Flash: 384K output), thinking mode, context caching; V4 Pro tops the Artificial Analysis Index among open weights (llm-benchmarks).
• DeepSeek R1 — MIT-licensed reasoning model, MoE 671B total / 37B active, 128K context.
• API: https://api.deepseek.com, OpenAI/Anthropic-SDK-compatible, integrates with Claude Code / GitHub Copilot (deepseek-api-docs); deprecated deepseek-chat/deepseek-reasoner retire 2026-07-24.

Why it matters to the landscape

DeepSeek is the clearest case of the open-weight-models + low-price one-two punch: by shipping capable open weights and an OpenAI-compatible budget API, it pulls the pricing floor down (llm-api-pricing) and pressures proprietary labs to justify their premium on capability (llm-benchmarks). Its API-compatibility strategy (conforming to incumbents’ shapes) lowers switching costs — a recurring challenger move.

Related

deepseek-api-docs · llm-provider · open-weight-models · llm-api-pricing · llm-benchmarks