Model Provider

DeepSeek

DeepSeek is a Chinese research lab known for cost-efficient Mixture-of-Experts models. Its open-weight V3 and R1 reasoning models match or exceed many closed frontier models on coding and math.

Visit DeepSeek

Models tracked

Open weight

API only

Avg score

59.8

Top benchmark

98.3

MATH-500

Total HF downloads

14.5M

Context window

128K – 1.0M

First release

Dec 2024

Latest release

Apr 2026

Top DeepSeek Models

Ranked by composite score across benchmarks, popularity, efficiency, and versatility.

#	Model	Modality	Score	Params	Released
1	DeepSeek-R1	text	BB66.6	671B	Jan 2025
2	DeepSeek-V4-Flash

Leaders by Modality

The strongest model from this provider in each modality. Click a card to open the model page.

Text

DeepSeek-R1

66.6

Modality Breakdown

How this provider’s catalog splits across text, image, video, audio, and embedding.

Text6 (100%)

Release Timeline

When each model shipped, newest first.

2026
- DeepSeek-V4-Pro
  Apr 24
  62.2

Open-Weight DeepSeek Models

Models with downloadable weights, ranked by composite score.

#	Model	Modality	Score	Params	Released
1	DeepSeek-R1	text	BB66.6	671B	Jan 2025
2	DeepSeek-V4-Flash

Model Families From This Provider

Each family is a model series. Browse all releases in a family with their own leaderboard.

DeepSeek

6 models in this family

Frequently Asked Questions

What models does DeepSeek ship?

DeepSeek ships 6 models that we track. Of those, 6 are open weight and 0 are accessed through their API.

Are DeepSeek models open source?

It depends on the model. See the open-weight section above for models you can run locally, and the API-only section for closed models priced per token.

Which DeepSeek model is best?

By composite score, DeepSeek-R1 is currently the top DeepSeek model in our directory. Composite score blends benchmarks, popularity, efficiency, and versatility.