In detail
- Models span Eagle3, DFlash, and DSpark series with parameters from 4B to 14B, based on Qwen3 and Gemma4 architectures.
- All models released as open weights on Hugging Face and are freely available.
- Releases occurred within minutes on the same day (12:25–12:37 UTC).
Why it matters
DeepSeek's continuous release of open-weight models in different sizes strengthens the supply of cost-effective, locally deployable AI solutions—relevant for companies seeking to reduce dependence on proprietary APIs.
For you Check whether these new DeepSeek models are cheaper for your inference needs than your current solutions, especially if you already work with Qwen or Gemma.