DeepSeek

DeepSeek V4 Flash

deepseek/deepseek-v4-flash-0731

·DeepSeek

DeepSeek V4 Flash is a 284B mixture-of-experts model activating roughly 13B parameters per token, with a 1M context window. It uses compressed latent attention, so a very long prompt costs far less memory than its length suggests. Built for work that needs a whole corpus in one pass — repository-wide code analysis, long document synthesis and multi-step reasoning over large inputs.

DeepSeek V4.1 Flash

deepseek/deepseek-v4.1-flash

·DeepSeek

DeepSeek V4.1 Flash is a sparse mixture-of-experts model and the first built on DeepSeek's Causal Encoder-Decoder architecture, activating 8B parameters on input and 16B on output. It accepts text and images, holds a 1,048,576-token context, and supports tool calling, structured output and reasoning control. The asymmetric activation makes it unusually cheap on long inputs, which suits whole-repository code work, long document synthesis and agent runs that accumulate a large history.

DeepSeek V4 Pro

deepseek/deepseek-v4-pro-0813

·DeepSeek

The GA release of DeepSeek V4 Pro: a large mixture-of-experts model with a one-million-token context window and strong tool-calling and agentic behaviour.