Google introduced Gemini 3, and immediately after its launch, the model caused a real stir in the AI community. Within the first 24 hours, over a million users tested it in Google AI Studio and through the Gemini API. The company integrated Gemini 3 into Google Search from day one, and on the LMArena platform, it quickly topped the rankings among leading models. This release received public congratulations from Sam Altman of OpenAI, Elon Musk of xAI, and Marc Benioff of Salesforce, who called Gemini 3 a “crazy leap” in speed, image quality, video, and text.
LMArena head Wei-Lin Chiang noted that Gemini 3 Pro has a clear advantage in coding, mathematics, and creative writing tasks, and its capabilities in agent programming often surpass Claude 4.5 and GPT-5.1. According to experts, on the ARC-AGI-2 test, it nearly doubled the performance of OpenAI GPT-5 Pro while performing tasks ten times cheaper. On SimpleQA, which covers simple questions from various fields, Gemini 3 Pro also outperformed competitors.
Professionals who work with AI daily note the confident results of Gemini 3 in various fields. Joel Hron from Thomson Reuters reported that the model showed significant progress in internal tests for legal and tax tasks, surpassing previous versions and some products from Anthropic and OpenAI.
At the same time, users from specialized fields, such as radiology, point out the difficulties of Gemini 3 with rare cases and specific tasks, where older models with individual training still remain better.
Some experts note that the new model does not always replace their main work tools yet. Some users, for example, still choose Claude for coding and ChatGPT for information retrieval. Tim Dettmers from Carnegie Mellon University noted that Gemini 3 sometimes does not clearly follow instructions, although overall it makes an impression as a strong universal model.
