DeepSeek

What it is. DeepSeek is the Chinese AI company behind the DeepSeek-V4 family of open-weight models and the low-cost DeepSeek API platform. Why it matters. DeepSeek set the price floor for the Chinese API market — its flash tier lists $0.15/1M input tokens — and publishes frontier models under MIT, the most permissive license in the ecosystem. Key characteristics. Current API lineup is DeepSeek-V4.1-Flash and DeepSeek-V4-Pro, both with 1M-token context; OpenAI-, Responses- and Anthropic-compatible endpoints; peak/off-peak pricing. What a professional should know. V4-Pro is marked deprecated with conflicting official pages on whether its API continues after 2026-09-14 — resolve that against the current change log before committing production workloads.

DeepSeek is a Chinese AI company (杭州深度求索人工智能基础技术研究有限公司) known for open-weight models and low-cost APIs. Its current API lineup is the DeepSeek-V4.1 family: deepseek-flash (DeepSeek-V4.1-Flash, released 2026-09-10, multimodal, MIT open weights) and deepseek-v4-pro (DeepSeek-V4-Pro-0813), both with 1M-token context windows and 384K maximum output.

The API platform (platform.deepseek.com) offers OpenAI-compatible, Responses and Anthropic-compatible endpoints with automatic KV caching and peak/off-peak pricing in USD. The founding date is not stated on the official pages fetched.

Why it matters

China AI Hub analysis: DeepSeek is the ecosystem’s price-and-openness reference point — MIT open weights (V4.1-Flash, V4-Pro, V3.2) plus the lowest budget-tier pricing ($0.15/1M input) make it the default anchor against which every other vendor’s cost and license are measured, and its open inference infrastructure (FlashMLA, DeepGEMM, DeepEP, 3FS) extends that influence below the model layer, as the AI infrastructure page details. China AI Hub analysis indicates DeepSeek’s structural role is the cost-and-openness floor — its 49B-active MoE design and off-peak discounts shaped the market’s pricing expectations, and its MIT releases shaped its licensing expectations, even as the V4-Pro deprecation ambiguity shows how quickly that frontier layer now turns over. China AI Hub analysis indicates the V4-Pro deprecation ambiguity is itself a structural signal — DeepSeek’s floor-setting now moves fast enough that even its own official pages disagree on whether a frontier tier still runs, the cost of being the market’s pace-setter. DeepSeek’s models also circulate through third-party surfaces: Coze’s model node routes to DeepSeek, and the model-neutral platform Dify runs DeepSeek weights alongside others.

Entity hub

Models

• DeepSeek-V4.1-Flash
• DeepSeek-V4-Pro
• DeepSeek-V3.2

Products

• DeepSeek API platform

API

• DeepSeek API

Agents

• DeepSeek Harness

Research / Technology

• AI Infrastructure
• Distillation
• Inference
• Mixture of Experts
• Quantization
• Reasoning Models
• How DeepSeek Changed China’s AI Market

Comparisons

• DeepSeek-V4.1-Flash vs GLM-5.3-Flash
• DeepSeek-V4-Pro vs Kimi K3
• DeepSeek-V4-Pro vs Qwen3.8-Max

Pricing

• DeepSeek pricing

Benchmarks

• AutomationBench
• CyberGym
• DeepSWE
• GPQA Diamond
• HLE
• Terminal-Bench

What is uncertain

• The founding date is not stated on the official pages fetched.
• V4-Pro’s deprecation is conflicting across official pages (news page vs change log).

Sources

evidence_id	source_name	source_url	source_type	published	verified	confidence	conflict
src-companies-deepseek-1	DeepSeek API docs — Models & Pricing	https://api-docs.deepseek.com/quick_start/pricing	Official documentation	—	2026-09-20	high	—
src-companies-deepseek-2	DeepSeek API Change Log	https://api-docs.deepseek.com/updates	Official documentation	—	2026-09-20	high	—
src-companies-deepseek-3	DeepSeek official site (EN)	https://www.deepseek.com/en/	Official	—	2026-09-20	high	—
src-companies-deepseek-4	DeepSeek Transparency Center	https://www.deepseek.com/en/transparency/	Official	—	2026-09-20	high	—

Labels used above: Official fact (from DeepSeek API docs and the official site), Vendor-reported claim (pricing and model capabilities published by DeepSeek), and China AI Hub analysis (our synthesis, always introduced as such).