DeepSeek
@deepseek-ai • Open source projects from deepseek-ai. Use this route to separate flagship concentration from portfolio breadth before you treat a publisher as broadly strong.
Portfolio concentration
79%
Top three share
Shows whether the organization is driven by one breakout repo or several visible projects.
Breadth
30 repos
Visible snapshot
14 repositories updated in the last 90 days.
Leading language
Python
Portfolio mix
Python (12), C++ (7), Unknown (5)
Average size
18.6K
Stars per repository
Useful for distinguishing one flagship-heavy publisher from a repeatable portfolio.
Updated: 2026-09-30(10d ago)GitHub API fallback30 repositories
79%
of the visible star count comes from this organization's top three repositories.
18.6K
stars per repository in this same snapshot.
Python
is the most common language here, with 14 repositories updated in the last 90 days.
Why this rank
This organization stands out because its public portfolio is relatively balanced across 30 repositories.
Balanced portfolio across 30 reposTop 3 share 79%
Organization pages work best when you separate portfolio breadth from flagship concentration. In DeepSeek's case, the visible top three repositories account for about 79% of total stars in this snapshot, which helps explain whether the organization is known for one breakout project or for a broader repeatable portfolio.
The dominant language mix here is Python (12), C++ (7), Unknown (5). That makes this page useful not just for popularity checks, but also for seeing what technical shape an organization's public ecosystem actually has.
Top Repositories
| # | Repository | Language | Stars |
|---|---|---|---|
| 1 | deepseek-ai/deepseek-harness DeepSeek Harness: Everything is a Plugin. | TypeScript | 246.9K |
| 2 | deepseek-ai/DeepSeek-V3 | Python | 104.5K |
| 3 | deepseek-ai/DeepSeek-R1 | 91.9K | |
| 4 | deepseek-ai/DeepSeek-OCR Contexts Optical Compression | Python | 23.9K |
| 5 | deepseek-ai/FlashMLA FlashMLA: Efficient Multi-head Latent Attention Kernels | C++ | 13.1K |
| 6 | deepseek-ai/3FS A high-performance distributed file system designed to address the challenges of AI training and inference workloads. | C++ | 10.3K |
| 7 | deepseek-ai/DeepEP DeepEP: an efficient expert-parallel communication library | Cuda | 10.3K |
| 8 | deepseek-ai/DeepGEMM DeepGEMM: clean and efficient BLAS kernel library on GPU | Cuda | 8.9K |
| 9 | deepseek-ai/open-infra-index Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation | 8.1K | |
| 10 | deepseek-ai/DeepSpec DeepSpec: a full-stack codebase for training and evaluating speculative decoding algorithms | Python | 7.2K |
| 11 | deepseek-ai/awesome-deepseek-agent | 6.2K | |
| 12 | deepseek-ai/smallpond A lightweight data processing framework built on DuckDB and 3FS. | Python | 5K |
| 13 | deepseek-ai/Engram Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models | Python | 4.7K |
| 14 | deepseek-ai/DeepSeek-OCR-2 Visual Causal Flow | Python | 3.5K |
| 15 | deepseek-ai/DualPipe A bidirectional pipeline parallelism algorithm for computation-communication overlap in DeepSeek V3/R1 training. | Python | 3K |
| 16 | deepseek-ai/TileKernels A kernel library written in tilelang | Python | 1.9K |
| 17 | deepseek-ai/DeepSeek-V3.2-Exp | Python | 1.7K |
| 18 | deepseek-ai/DeepSeek-Math-V2 | Python | 1.6K |
| 19 | deepseek-ai/EPLB Expert Parallelism Load Balancer | Python | 1.4K |
| 20 | deepseek-ai/DeepSeek-Prover-V2 | 1.3K | |
| 21 | deepseek-ai/profile-data Analyze computation-communication overlap in V3/R1. | 1.2K | |
| 22 | deepseek-ai/DeepGEMM-Ascend DeepGEMM-Ascend: clean and efficient matrix multiplication kernel library for Huawei Ascend NPUs | C++ | 563 |
| 23 | deepseek-ai/LPLB An early research stage expert-parallel load balancer for MoE models based on linear programming. | Python | 538 |
| 24 | deepseek-ai/DeepSelect DeepSelect: TopK kernels for DeepSeek Sparse Attention (DSA) and Samplers | Cuda | 485 |
| 25 | deepseek-ai/deepseek-recipe | Rust | 381 |
| 26 | deepseek-ai/DeepJIT A lightweight library for xPU kernel JIT compilation | C++ | 379 |
| 27 | deepseek-ai/DeepEP-Ascend A high-performance communication library for machine learning training and inference on Huawei Ascend NPUs. | C++ | 248 |
| 28 | deepseek-ai/dsh-libreoffice-kit An internal component used by DeepSeek Harness | JavaScript | 169 |
| 29 | deepseek-ai/clangd-ascend Clangd for Ascend for Code Lint & Code Completion | C++ | 19 |
| 30 | deepseek-ai/dsh-node-addon-require-builtin An internal component used by DeepSeek Harness | C++ | 14 |
Next step after the organization read
Open a flagship repository, compare a couple of portfolio leaders, or return to the organization map when you want a broader concentration read.
Learn and methodology
Keep trust-building context reachable, but behind the first data read instead of ahead of it.
How to read this organization snapshot
Total stars are useful as a discovery signal, but they do not tell you whether a team maintains every repository equally. Pair this page with release cadence, maintainer activity, and the flagship concentration shown above before making adoption decisions.
For broader background on GitStar's ranking logic and editorial guidance, see Methodology & Editorial Standards.