New Release
DeepSeek-V4: a million-token context that agents can actually use
DeepSeek-V4 revolutionizes AI interactions with a million-token context for more complex conversations.
Hugging Face BlogOriginal →
DownloadableModel
DeepSeek (via Ollama)
DeepSeek
DeepSeek publishes open-weight models spanning small distilled tags to very large sizes.
Best for
Best for strong open-weight reasoning/coding locally; weak if you lack VRAM for mid-size tags.
Pricing
Free to use at the access path below — no per-token bill for this tier.
What it does well
Strong reasoning/coding among open-weight options.
Where it struggles
Full-size needs serious hardware; distill tags trade capability.
Practical fit
- Local reasoning-heavy prompts
- Offline coding help
Where to run it
Runs locally via Ollama — no per-token billing.
Install & setup
1. Install Ollama. 2. Run: ollama pull deepseek-r1 3. Prefer a distilled tag (e.g. deepseek-r1:7b) on limited hardware.
What you need to run it
Minimum: 8GB RAM for smallest distill. Mid tags: 16GB+ VRAM recommended.
License & download
License: Model-specific (check release notes)
Suggested VRAM: ~8 GB
Download: https://ollama.com/library/deepseek-r1
Official: ollama.com/library/deepseek-r1
Share
Related coverage
Developing
The Future of the Global Open-Source AI Ecosystem: From DeepSeek to AI+
The open-source AI ecosystem is evolving, with DeepSeek paving the way for AI+ to enhance collaboration and innovation.
Hugging Face BlogOriginal →
New Release
Architectural Choices in China's Open-Source AI Ecosystem: Building Beyond DeepSeek
China's open-source AI ecosystem is evolving with innovative frameworks that promise enhanced efficiency and scalability.
Hugging Face BlogOriginal →