DeepSeek New Model Enters Beta Testing! Users Report Blazing Speed

On September 8, an intermediate version of DeepSeek V4.1 Flash began beta testing, featuring a new model architecture with native multimodal support, emphasizing stronger capabilities, faster speed, and lower costs, with pricing on par with V4 Flash. Users reported output speeds reaching up to 507 tokens/s in real-world tests, potentially signaling the start of a new iteration cycle for DeepSeek's models.