Google Gemini 3 Pro Beats GPT‑5.1 on Top AGI Benchmarks – What the Results Reveal

Google unveiled Gemini 3 and the Gemini 3 Pro model, positioning them as a step forward on the path to artificial general intelligence (AGI). The announcement highlighted the models' performance on several high‑profile leaderboards.

The Gemini models claimed the number‑one spot on the Arena leaderboard and dominated almost every sub‑ranking, especially the two most demanding benchmarks: The Human Last Exam and ARC‑AGI‑2 . These tests are designed to evaluate general intelligence rather than narrow, task‑specific abilities.

The Human Last Exam is a multimodal benchmark containing roughly 3,000 extremely difficult questions. It is described as a "Turing test for large models," comparable to MMLU (a high‑school‑level exam) and AIME (a competition‑level math test), but far more comprehensive.

ARC‑AGI originated from the creator of Keras and features abstract reasoning problems similar to civil‑service exam graphics. ARC‑2 pushes the difficulty even further, making it a stringent test of a model's reasoning capabilities.

To illustrate the practical gap, the article presented a prompt:

Create the best visualization of a spur gear in 3D possible, without external libraries. It should be fully math‑based and include a stress analysis and contact analysis.

GPT‑5.1 produced a result after about seven minutes, which was described as coarse. Gemini 3 completed the same task in roughly thirty seconds, delivering a far more detailed gear shape, realistic lighting, and perspective.

Another striking example asked the model to build a cloud‑based operating system supporting a graphical UI, text editor, web browser, and real‑network access, with Wikipedia set as the homepage. Gemini 3 generated a functional OS environment where the built‑in browser could actually load Wikipedia, demonstrating a level of integration previously unseen.

The piece concludes that the large‑model race has moved beyond merely climbing leaderboards. Leading companies are now laying down foundational research that could pave the way to true AGI, attracting top talent to foundational R&D rather than traditional product development.