DeepSeek Web Version Launches 'Quick' and 'Expert' Dual Modes with Product Layering

According to monitoring by 1M AI News, the DeepSeek web version has introduced a mode-switching feature, with lightning and diamond icons appearing above the input box, corresponding to ‘Quick Mode’ and ‘Expert Mode’, respectively. Quick Mode is suitable for everyday conversations, providing instant responses and supporting image and file recognition; Expert Mode is aimed at complex reasoning tasks but currently does not support file uploads or multimodal capabilities. The update was launched silently without any official announcement. Community analysis reveals that Quick Mode operates on DeepSeek 3.2, with a knowledge cutoff date of July 2024; Expert Mode points to a newer model, possibly an early version of V4. User tests indicate that Expert Mode performs significantly better in deep reasoning tasks such as physical simulations and mathematical reasoning, but shows limited differences from Quick Mode in simpler tasks like creative writing. Some testers believe that Expert Mode is still routed through a version of V4 Lite, with the full version of V4 yet to be released. The frontend code also reveals an upcoming third option, ‘Vision Mode’, but reverse analysis shows it is not an independent model; rather, it activates a visual understanding parameter within Quick Mode. This marks the first instance of product layering for DeepSeek since its rise to prominence early last year. Previously, the web version was completely free and without functional distinctions, directing users to different model entrances based on their needs, which itself was a computational resource scheduling strategy. Once this architecture is established, there will be no technical barriers to introducing a paid system or limiting access by mode in the future.