On the evening of April 7, 2026, DeepSeek's web interface quietly received an update, introducing a new 'Quick Mode' and 'Expert Mode'. Quick Mode is ideal for daily conversations, offering instant responses and supporting image and file text recognition, likely powered by a lightweight optimized V4 Lite model. Expert Mode excels at handling complex problems, though it may require waiting during peak times. It is suspected to be connected to the prototype of the official V4 version, demonstrating superior performance in deep tasks such as physical simulations and mathematical reasoning. Currently, the new version is in grayscale testing and is not yet fully rolled out.
