The History of AI

Gemini 3 Pro

Era 5 · Generative AI, reasoning, and agents (2022–2026) › Frontier models

Lineage
Neurons and networks
Kind
Implementation

Who

Google DeepMind

What happened

Google presents Gemini 3 and releases its first model, Gemini 3 Pro, which tops the LMArena leaderboard. For the first time, a Gemini model reaches Google Search on its launch day.

Why it mattered

According to Google, it beats its predecessor on every major benchmark. It reaches Search from day one.

The full story

On November 18, 2025, Sundar Pichai, Demis Hassabis, and Koray Kavukcuoglu, chief technology officer of Google DeepMind, presented Gemini 3. Its first model, Gemini 3 Pro, came out that day as a preview.

It is a multimodal reasoning model: it takes text, images, and audio, accepts up to a million input tokens, and uses a sparse mixture of experts, an architecture that activates only part of the network for each response.

Google supplied figures. Gemini 3 Pro topped LMArena with 1,501 Elo points, and it scored 37.5 percent on Humanity's Last Exam, without tools, and 76.2 percent on SWE-bench Verified, which measures the resolution of real programming issues. According to the company, it clearly beat Gemini 2.5 Pro on every major benchmark.

The rollout changed too. The model reached the Gemini app and developers that day and, for subscribers, AI Mode in Search: it was the first time Google had put a Gemini model into Search on its launch day. Alongside it the company presented Antigravity, an agent-based coding platform. Deep Think mode, with longer reasoning, was held back pending further safety evaluations.

Six days later Anthropic released Claude Opus 4.5.

Sources

Scholarship and context

How to cite

“Gemini 3 Pro.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/gemini-3-pro

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline