DeepSeek V4.1 Flash

DeepSeek V4.1 Flash is a sparse mixture-of-experts model from DeepSeek and the first built on its Causal Encoder-Decoder design. It was listed on OpenRouter on 10 September 2026. The context window is 1,048,576 tokens, it reads text and images, and it writes text. On 5 November 2026 the price is $0.045 per million input tokens and $0.50 per million output, among the lowest in this batch. The weights are open: the Hugging Face card says the repository and model weights are under the MIT licence, which allows commercial use. The card gives 8B active parameters during prefill and 16B during decode, plus a large sparse memory store, and lists two different total-size figures without explaining them. The limit is that output is text only. It suits cheap, long-context work and anyone who wants to self-host under a permissive licence. You reach it through DeepSeek's API, OpenRouter, or other hosts. A dated sibling, deepseek-v4-1-flash-1017, is already on our list. Figures come from OpenRouter and the Hugging Face card. We have not run it.