Google overhauls Gemini: thinking levels replace the Thinking model

Google has completely overhauled Gemini. Both the web version and the smartphone app were updated, with a new interface, built-in thinking-depth modes and a new lightweight model, Flash Lite. The update is big and highly visible, but it brings both gains and losses. Users lost the standalone Thinking model and now face stricter limits on the Pro subscription.

Three levels instead of a standalone Thinking model

The main change is architectural. Previously, Thinking was a separate model that you had to pick from a menu. Now Google has built thinking-level selection straight into the main chat interface: three buttons for any question.

• Fast — instant answers, on-the-fly analysis, no deep reasoning
• Balanced — the standard mode, best for most everyday tasks
• Deep — extended reasoning for complex questions, analytics and creative tasks

That sounds more convenient. You no longer have to hunt through a menu: you press a button in the chat and type your query. You get results faster than when looking for a separate model.

But there is a catch. Deep mode is not a full replacement for the Thinking model. It runs noticeably faster than Thinking and uses fewer compute resources, but it also analyzes less deeply. It thinks, just not for as long. For complex scientific questions, financial modeling and long code, Thinking was stronger. Deep falls short.

Flash Lite and the redesigned interface

The new Flash Lite model is built for quick queries: translating between languages, summarizing texts, looking up factual information, and simple questions like "when...", "where...", "how many...". It responds almost instantly because it is specifically optimized for speed. It uses fewer tokens and gives shorter answers, though it is most accurate on simple tasks.

Gemini's entire interface was rebuilt from scratch. Unnecessary visual elements were removed and buttons were placed more intuitively. On mobile it is now easier to use: the mode selector is clearly visible, with no hidden menus. The desktop version was redesigned too, and is cleaner and more minimalist.

The losses: Thinking is gone, Pro is restricted

There are serious losses as well. The Thinking model was removed entirely. Those who had started actively using it over the past couple of weeks were suddenly left without the tool. Deep mode does not replace it: it works more simply and quickly, but is less analytical.

The Pro subscription got stricter limits. Subscribers used to be able to send more requests per day. Now the ceiling is lower. Google has not given exact figures, but Reddit and other forums are full of complaints. People who pay for the product got fewer capabilities.

Free users got access to the modes (Fast, Balanced, Deep), but with a cap on the number of Deep requests per day. The result is odd: you pay for the product, you pay for the limits, and you still lose features.

What this shows

Google is clearly changing strategy. Instead of separate specialized models, the company is building options directly into the main product. That is more convenient in the interface, but less flexible for experienced users who need deep tools.

The Pro subscription is becoming less attractive. Limits are growing, functionality is getting duller, and the best features increasingly show up in the free version. It is a typical commercial dynamic: first you hand out features generously, then you start restricting them.

For OpenAI, this is a window of opportunity. Its o1 and o3 are still more powerful at deep analysis, and while Google is ticking boxes in the interface, competitors can focus on quality.