Loading...

You don't need another chatbot. You need proof you can build something no one's built before. Every model, every agent, every experiment. Your table. You set it.

Platform

Knowledge
Visual Workflows
Collaboration Engine
Model Catalog

Resources

Research
Datasets

Legal

Terms of Service
Privacy Policy

© 2026 Hermosa Labs, LLC. All rights reserved.

Built in California 🌊🌊🌊

DeepSeek

[Hangzhou, China][founded 2023][backed by High-Flyer; $7B Series A (May 2026)]www.deepseek.com

DeepSeek is a Chinese AI lab founded in July 2023 in Hangzhou by Liang Wenfeng. It was funded solely by High-Flyer, the quantitative hedge fund Liang co-founded, until a $7 billion Series A in May 2026, and grew out of GPU capacity originally stockpiled for trading research. The lab releases its V, R and V4 model lines as open weights under the MIT licence, with the V4 Pro and Flash tiers succeeded by V4.1 Flash in September 2026.

sources:en.wikipedia.org api-docs.deepseek.com/…/news260424 api-docs.deepseek.com/…/news260910

· verified 2026-09-13

8

models

3

families

8

open weights

8

3d teardowns

2023–2026

shipping

■ text 8

8 models · 3 families · 2023–2026

V series (base and chat)

newest: DeepSeek V3.2 · 2025-12-01

[frontier][open-weights][api]

R series (reasoning)

newest: DeepSeek R1 (0528) · 2025-05-28

[reasoning][open-weights][api]

V4 (Pro and Flash tiers)

newest: DeepSeek V4.1 Flash · 2026-09-10

[frontier][open-weights][api][Pro / Flash]

Families (3)

V series (base and chat)

[frontier][open-weights][api][superseded]

2 in catalog · 3 earlier

before the catalog2023-11 DeepSeek LLM 67B, dense, first release→2024-05 DeepSeek V2, first MoE with MLA→2024-12 DeepSeek V3, FP8 training, MTP→

DeepSeek V3 (0324)

code and maths bump, MIT licence

[text]164k ctx[open-weights · MIT][3d teardown]

DeepSeek V3.2

sparse attention

[text]164k ctx[open-weights · MIT][3d teardown]newest

sources:en.wikipedia.org api-docs.deepseek.com

R series (reasoning)

[reasoning][open-weights][api][superseded]

1 in catalog · 1 earlier

before the catalog2025-01 DeepSeek R1, RL-trained reasoning on the V3 base→

DeepSeek R1 (0528)

May 2025 checkpoint refresh

[text]164k ctx[open-weights · MIT][3d teardown]newest

sources:en.wikipedia.org api-docs.deepseek.com

V4 (Pro and Flash tiers)

[frontier][open-weights][api][Pro / Flash]

5 in catalog

DeepSeek V4 Pro 0423 (preview)

1.6T MoE (49B active), 1M context, preview

[text]1M ctx[open-weights · MIT][3d teardown]

DeepSeek V4 Flash 0423 (preview)

284B MoE (13B active), 1M context, preview

[text]1M ctx[open-weights · MIT][3d teardown]

DeepSeek V4 Flash 0731

official Flash release, 304B, DSpark speculative decoding

[text]1.3M ctx[open-weights · MIT][3d teardown]

DeepSeek V4 Pro 0813

official Pro release, 1.7T, reasoning effort levels

[text]1M ctx[open-weights · MIT][3d teardown]

DeepSeek V4.1 Flash

552B MoE, causal encoder-decoder, multimodal; replaces V4 Flash

[text]1M ctx[open-weights · MIT][3d teardown]newest

sources:api-docs.deepseek.com/…/news260424 api-docs.deepseek.com/…/news260813 api-docs.deepseek.com/…/news260910