“We’ll build the first one.” — Sergey Brin’s surprise return at Google I/O 2025 came with a bold claim: Google is officially aiming to develop the world’s first artificial general intelligence (AGI).
This marks a major shift in tone, with Google now openly entering the AGI race alongside OpenAI, Anthropic, and xAI. Brin’s remarks follow the launch of Gemini 2.5, which, while more advanced, still falls short of true AGI.
The message is clear: Google’s founders are back in the driver’s seat—and the stakes just got higher.
Latest comparisons of the models
Gemini 2.5 Pro vs. GPT-4o:
- Context Window (Memory): Gemini 2.5 Pro wins BIG. It handles 1 million tokens (and soon 2 million), letting it analyze entire books or huge codebases. GPT-4o is limited to 128,000 tokens.
- Reasoning/Problem Solving: Gemini 2.5 Pro often leads, especially in complex science and math (like the GPQA Diamond and AIME benchmarks) and in its unique “Humanity’s Last Exam.” GPT-4o is still very strong in general knowledge (MMLU).
- Multimodality (Images, Video, Audio): Gemini 2.5 Pro shows stronger understanding on benchmarks like MMMU. It’s built from the ground up for this. GPT-4o is also excellent, especially for real-time voice interactions.
- Code Generation/Editing: Gemini 2.5 Pro has an edge in complex software engineering tasks and editing, as seen in SWE-Bench and Aider Polyglot. GPT-4o is still very good for general code generation.
- Cost/Speed: Gemini 2.5 Pro (and especially Gemini 2.5 Flash) can be significantly more cost-effective and faster for high-volume use cases compared to GPT-4o.
In a nutshell: Gemini 2.5 Pro is often ahead for deep analysis of huge amounts of data, complex reasoning, and advanced coding. GPT-4o remains incredibly versatile and a top choice for broad tasks and quick, seamless voice interactions.
