DeepSeek V4 Pro DSpark is a text-in, text-out open-weight model from deepseek-ai with a very large context window of roughly one million tokens — meaning it can hold and reason over extremely long documents in a single pass. Details about its specific capabilities and what distinguishes it from other DeepSeek models are limited based on available information. It ships in safetensors format under an MIT license, making it freely usable and modifiable.