According to CoinMeta, DeepSeek has released the weights for its first visual model, deepseek-v4-flash-vision-exp. This experimental visual model, based on v4-flash, has added the ability to understand images. It can analyze screenshots and read charts, and can also be used in conjunction with tools to complete agent tasks. In pure-text agent tasks, it still performs similarly to v4-flash. Previously, this model could only be used through API; now, developers can download the weights and deploy it on their own.