Product line release timeline
Generational evolution by product line · dot = one model release · dashed line connects successive generations · click a dot to open the model page
Published models
40 modelsModels published by DeepSeek-AI, grouped into 17 series.
DeepSeek V4
- First
- 2026-04-24
- Latest
- 2026-04-24
- Span
- 1 days
- Models
- 4
DeepSeek OCR 2
- First
- 2026-01-28
- Latest
- 2026-01-28
- Span
- 1 days
- Models
- 1
DeepSeek V3.2
- First
- 2025-09-29
- Latest
- 2025-12-01
- Span
- 64 days
- Models
- 3
DeepSeekMath V2
- First
- 2025-11-27
- Latest
- 2025-11-27
- Span
- 1 days
- Models
- 1
DeepSeek OCR
- First
- 2025-10-20
- Latest
- 2025-10-20
- Span
- 1 days
- Models
- 1
DeepSeek V3.1
- First
- 2025-08-20
- Latest
- 2025-09-22
- Span
- 34 days
- Models
- 2
DeepSeek R1 Distill
- First
- 2025-01-20
- Latest
- 2025-05-30
- Span
- 131 days
- Models
- 3
DeepSeek R1
- First
- 2025-01-20
- Latest
- 2025-05-28
- Span
- 129 days
- Models
- 3
DeepSeek Prover V2
- First
- 2025-04-30
- Latest
- 2025-04-30
- Span
- 1 days
- Models
- 1
DeepSeek V3
- First
- 2024-12-26
- Latest
- 2025-03-24
- Span
- 89 days
- Models
- 3
DeepSeek R1 Lite
- First
- 2024-11-20
- Latest
- 2024-11-20
- Span
- 1 days
- Models
- 1
DeepSeek V2.5
- First
- 2024-09-05
- Latest
- 2024-09-05
- Span
- 1 days
- Models
- 1
DeepSeek V2
- First
- 2024-05-06
- Latest
- 2024-05-06
- Span
- 1 days
- Models
- 2
DeepSeek VL
- First
- 2024-03-11
- Latest
- 2024-03-11
- Span
- 1 days
- Models
- 4
DeepSeekMoE
- First
- 2024-01-11
- Latest
- 2024-01-11
- Span
- 1 days
- Models
- 4
DeepSeek LLM
- First
- 2023-11-29
- Latest
- 2023-11-29
- Span
- 1 days
- Models
- 4
DeepSeek Coder
- First
- 2023-11-01
- Latest
- 2023-11-01
- Span
- 1 days
- Models
- 2
About this organization
Organization Overview
DeepSeek is China's leading large-scale model company. It was incubated by High-Flyer, a well-known quantitative private equity company. It is headquartered in Hangzhou and is committed to exploring the nature of general artificial intelligence. The company is known as open source, and its models and codes can be freely downloaded, modified and used. It has had a wide impact on a global scale with its architectural innovation and extreme training cost control.
Creation and Development
-
Established in July 2023, relying on Magic Square Quantization’s deep accumulation in GPU computing power and engineering, led by Liang Wenfeng.
-
December 2024 open source DeepSeek-V3, training results comparable to the world's top closed-source models at a cost far lower than the industry.
-
January 2025 The large open source inference model DeepSeek-R1 has performance close to OpenAI o1 and is completely open source. Its AI assistant once topped the App Store in many countries, attracting global attention and the "DeepSeek moment".
Flagship model
-
General/Inference: The latest flagship is DeepSeek-V4 (including V4-Pro, V4-Flash), which continues the efficient MoE architecture of the V3.1 / V3.2 series
-
Mathematics/Proof: DeepSeekMath-V2, DeepSeek-Prover-V2
-
Multi-modal/OCR: DeepSeek-VL series, DeepSeek-OCR/OCR 2
-
Programming: DeepSeek-Coder series
Latest Development (2026)
-
Launched the company's first round of large-scale external financing, with a scale of approximately US$7 billion (approximately RMB 50 billion), and a valuation of approximately US$52-59 billion.
-
Investors include Tencent, CATL, JD.com, etc. Founder Liang Wenfeng’s personal subscription accounted for approximately 40% of this round.
Leader: Liang Wenfeng (Founder) Official website: deepseek.com
Open source homepage: github.com/deepseek-ai Online assistant: chat.deepseek.com
Note: The financing and valuation data in this section are public information as of mid-2026 and may change over time.