Google Launches Gemini 3

Google just rolled out Gemini 3, and it’s scoring higher on reasoning benchmarks than GPT 5 Pro.

The model hit a score of 37.4 on the Humanity’s Last Exam benchmark, designed to measure high-level reasoning and general expertise, compared to GPT 5 Pro’s 31.64. 

Google released it just weeks after OpenAI launched GPT 5.1 and Anthropic released Sonnet 4.5, continuing a pattern of increasingly rapid model releases across the industry.

What’s interesting is that the frontier has shifted from just building bigger models to optimizing for specific reasoning capabilities. Earlier this month, OpenAI launched GPT 5.1 which focuses more on reasoning than pure computing capacity. Gemini 3 comes seven months after Gemini 2.5, which itself came only months before that. The rollout includes guardrails designed to prevent the kinds of errors that plagued earlier models, though whether those actually hold when millions of people are using the system remains to be seen.

The model is live for Gemini Pro and Ultra subscribers in the US, with a more research-focused version called Gemini 3 Deepthink coming later. Google also introduced Antigravity, a coding interface where the AI agent can work across your editor, terminal, and browser simultaneously.

At this pace, the capability gap closes and reopens within weeks. We are in an era where AI companies are in a neck to neck competition.