Model by deepseek-ai

deepseek-v4.1-flash

552B MoE, 8B active params with native multimodal support and lower API cost using smaller KV cache