DeepSeek Web Version Quietly Launches "Fast" and "Expert" Dual Modes, Introducing Product Layering for the First Time

CoinJie.com news: the DeepSeek web app has added a mode-switching feature. Two icons—an electric bolt and a diamond—now appear above the input box, corresponding to “Quick Mode” and “Expert Mode,” respectively. Quick Mode is suited for everyday conversations, delivers instant responses, and supports image and file recognition. Expert Mode is aimed at complex reasoning tasks, but currently does not support file uploads or multimodal capabilities. The update went live completely silently, with no official announcement. Community analysis shows that Quick Mode is running DeepSeek 3.2 behind the scenes, with a knowledge cutoff date of July 2024; Expert Mode points to a newer model, possibly an early version of V4. In user tests, Expert Mode performs noticeably stronger on deep-reasoning tasks such as physical simulation and mathematical reasoning, but for simpler tasks like creative writing, the gap compared with Quick Mode is limited. Some testers believe the Expert Mode route is still a certain version of V4 Lite, and that the full V4 release will need to wait. In the front-end code, a third option called “Vision Mode” was also found that has not been launched yet, but reverse engineering indicates it is not an independent model; instead, a visual understanding parameter is enabled in Quick Mode. This is DeepSeek’s first time doing product tiering since it went viral in early last year. Previously, the web app was completely free with no feature differentiation, and it funneled users based on their needs to different model entry points—on its own, that is a kind of compute-scheduling strategy. And once this architecture is in place, there are no technical obstacles afterward, whether it’s introducing a paid system or limiting usage by mode quotas.