DeepSeek released a distilled model, DeepSeek-R1-0528-Qwen3-8B, obtained by distilling Qwen3-8B using R1-0528, which supports deep thinking and Function Calling and has been specifically optimized for MCP.