Google has unveiled a series of updates for its Gemini AI platform. The spotlight is on the experimental Gemini 2.0 Flash model featuring an advanced thinking mode, a new AI agent called Jules for coding, and several improvements for Gemini Advanced subscribers.
Gemini 2.0 Flash: More Time for Thinking
The main highlight is the experimental advanced thinking mode in Gemini 2.0 Flash. This is a version of the model that can think for a longer time before giving an answer. Simply put: instead of immediately returning a result, the model spends more time analyzing the task. This is especially useful for complex queries – programming, mathematics, and multi-step problems.
Google tested it on several standard benchmarks. On the AIME 2024 math test, it scored 71.9% – higher than the base version of Gemini 2.0 Flash (55.4%). Performance on programming tasks (Codeforces) also improved, rising from the 56th to the 86th percentile. On the graduate-level humanities test (Humanities Graduate Level Reasoning / GPQA Diamond), it scored 65.3% versus 51.1%.
The model is available via AI Studio and the Gemini API. It is currently in the testing phase, so results may be unstable. However, if you need a model for tasks where reasoning accuracy is crucial, it's worth trying.
Jules – An AI Agent for Coding
Google introduced Jules, an AI agent capable of working with code directly on GitHub. You can assign it a task, such as fixing a bug or preparing a pull request, and it will locate the necessary files, make changes, and create a work plan on its own.
Jules is based on Gemini 2.0 and integrates into GitHub workflows. You create an issue; Jules picks it up, analyzes the codebase, proposes a solution, and prepares the changes. The agent can also pause and ask for clarification if something is unclear. A developer can check what Jules is doing at any moment and adjust its actions.
For now, this is an experimental version available to a limited number of testers. Sign-ups for testing are open on the Google Labs website. Jules can help speed up routine tasks: code preparation, refactoring, and bug fixing. However, it won't replace a developer – it's an assistant that reduces some of the workload.
What Else Was Added to Gemini Advanced
There are updates for Gemini Advanced subscribers as well. The model can now handle a much larger context window – up to 1 million tokens. This means you can upload a large document, multiple files, or a long conversation history, and Gemini will be able to work with them without losing details.
Another new feature is integration with Google Calendar. Gemini can now create calendar events, check schedules, find free time, and even plan meetings based on your requests. For example, you can ask: «Find free time next week for a one-hour team meeting», and Gemini will suggest options.
Furthermore, the Gemini mobile app now allows sharing files and images directly from other apps via the standard «Share» menu. This simplifies workflows: there is no need to switch between apps; you can send a file to Gemini immediately and ask a question.
Google has also included a new version of the image generator – Imagen 3 – in Gemini Advanced. The model produces more realistic and detailed images, understands text prompts better, and more accurately renders lighting, textures, and fine details.
Imagen 3 is integrated into the Gemini interface, so images can be created directly within the chat. This is convenient if you need to quickly visualize an idea, prepare an illustration, or experiment with generation.
What This Means for Developers and Users
The updates show that Google continues to develop Gemini in two directions: improving the model's reasoning capabilities and creating specialized tools for specific tasks.
Gemini 2.0 Flash with its advanced thinking mode will be useful for those working on complex problems. If you need a model that doesn't just answer quickly but instead carefully analyzes the query, this version might come in handy. However, since it is experimental, its behavior is not always predictable.
Jules is an early attempt to create an AI agent that genuinely helps in development. It is too early to judge how well it handles real-world tasks, but the direction is clear: automating routine work to give developers more time for solving complex problems.
For regular users, the key features of Gemini Advanced are the expanded context window and calendar integration. These additions make the model more practical: you can work with large documents and plan tasks without switching between different tools.
Overall, the updates represent a step toward making AI not just a smart assistant, but a tool that integrates into daily workflows. Only time and user experience will tell how well this works in practice.
