Google Unleashes Gemini 3 Pro, Claiming AI Performance Crown

Google has officially launched its highly anticipated Gemini 3 Pro AI model. This release marks a significant leap in the company’s artificial intelligence capabilities. The new model is now rolling out globally to users.

According to Reuters, Google describes Gemini 3 as its most capable and natively multimodal reasoning model to date. The company positions this launch as a major step toward achieving Artificial General Intelligence (AGI).

Gemini 3 Pro Outperforms Rivals in Key Benchmarks

Gemini 3 Pro is built on a sophisticated sparse mixture-of-expert architecture. It was trained exclusively on Google’s custom Tensor Processing Units. This specialized training infrastructure contributed to its advanced performance.

The model has achieved top scores in several critical industry benchmarks. On the challenging “Humanity’s Last Exam,” Gemini 3 Pro scored 37.5% without using any external tools. This performance notably surpassed competing models from other leading AI labs.

In the new ARC-AGI-2 benchmark, Gemini 3 Pro achieved a score of 31.1%. It also leads the LMArena leaderboard with a top ELO rating. These results demonstrate its superior reasoning and problem-solving abilities across diverse tasks.

Advanced Capabilities and Wider Market Impact

Google is also developing an even more powerful version called Gemini 3 Deep Think. This variant has shown even higher scores in early testing. It is expected to be available soon for Google AI Ultra subscribers.

The model excels at agentic coding and long-horizon planning. It can break down complex, multi-step tasks and execute them effectively. This makes it particularly useful for software development and complex project management.

Google introduced a new developer platform named Antigravity alongside the model. This environment integrates several AI tools for a seamless development experience. It allows AI agents to directly control editors, terminals, and browsers.

Gemini 3 Pro’s release intensifies the competition in the frontier AI market. Its performance claims set a new bar for public AI model capabilities. This advancement puts pressure on other tech giants to accelerate their own AI development cycles.

The launch of Gemini 3 Pro solidifies Google’s position in the top tier of AI development. This model represents a tangible step toward more generalized and capable artificial intelligence. Its performance will likely shape the next phase of the global AI race.

Info at your fingertips

What is Google Gemini 3 Pro?

Google Gemini 3 Pro is a state-of-the-art AI model. It is a natively multimodal and reasoning model. Google claims it represents a major step toward advanced AI capabilities.

How does Gemini 3 Pro perform against competitors?

It outperforms other leading models in key benchmarks. It achieved a 37.5% score on Humanity’s Last Exam. It also leads in LMArena and ARC-AGI-2 evaluations.

When can users access Gemini 3 Pro?

The model is rolling out now in the Gemini app. It is available for all users starting today. Google AI Ultra subscribers get additional features in search.

What is the Antigravity platform?

Antigravity is a new agent-first development environment. It bundles Gemini 3 Pro with other specialized AI models. It allows agents to control coding tools directly.

What makes Gemini 3 Pro’s architecture unique?

It uses a sparse mixture-of-expert (MoE) design. It was trained solely on Google’s custom TPU hardware. This architecture contributes to its high efficiency and performance.

Trusted Sources
Reuters, Associated Press