DeepSeek-V4.1-Flash, the smallest model in DeepSeek's new V4.1 architecture family, is now available through DigitalOcean Inference Engine. A new Causal Encoder-Decoder architecture activates just 8B parameters on input and 16B on output, cutting KV cache size to roughly 1/4 the memory and 1/8 the storage of the prior Flash generation, while outperforming the much larger DeepSeek-V4-Pro on most agentic and coding benchmarks, under the MIT License.