The most notable highlight of Gemini 3.5 Pro is its support for a context window of up to 2 million tokens. This would allow the model to remember and process a very large amount of data in a single interaction, enabling it to maintain long conversations, analyze complex documents and provide more coherent responses. In addition, Google is also said to be adding a new reasoning mode called “Deep Thinking,” designed to improve logical reasoning and multi-step problem solving.

Gemini 3.5 Pro is reportedly designed to serve complex workflows for AI agents. The model will support action execution, multi-agent coordination, programming, multimodal content generation and long-duration task processing. This suggests that Google wants to develop an AI system that can not only answer questions but also proactively carry out complex chains of work, going far beyond the role of Gemini 3.5 Flash.

The source also shared a screenshot allegedly taken from the LMSYS Chatbot Arena platform. According to it, Gemini 3.5 Pro outperformed Anthropic’s Claude Fable 5 High model in a programming test. Specifically, Google’s model generated minimalist SVG graphics for a card reader device with better results.

In addition, two videos comparing UI-generation capabilities between Gemini 3.5 Pro and Claude Fable 5 also showed that Google’s model produced more attractive interface designs, better SVG graphics and smoother AI-agent workflows. If this information is accurate, Gemini 3.5 Pro could become a notable tool for developers and users who need an AI capable of creativity and automated handling of complex tasks.

At the moment, Google has not officially confirmed the above information, so all details about Gemini 3.5 Pro still need to await the company’s announcement.

Conclusion:
If the reported features are confirmed, Gemini 3.5 Pro could become a strong tool for developers and advanced AI-agent use cases. However, since Google has not officially confirmed the details, the model’s real performance and positioning should be assessed only after launch.