Google Releases Its Most Powerful AI Model, Gemini 3—Here's What You Need to Know

Google introduced Gemini 3 Pro, its most advanced AI model, which significantly outperforms its predecessor across reasoning and multimodal benchmarks. This model can process up to 1 million tokens of context, handling text, images, audio, and video. It was made available for public preview. The benchmarks show Gemini 3 Pro scoring 37.5% on an academic reasoning test, a substantial improvement over the 21.6% achieved by the previous Gemini 2.5 Pro. Despite these technical advancements, Google faces stiff competition in the AI market, particularly from OpenAI's ChatGPT, which boasts approximately 800 million weekly users compared to Gemini's 650 million monthly users. Gemini 3 Pro employs a sparse mixture-of-experts architecture, activating only the relevant parameters for specific tasks, which optimizes computational efficiency. Initial tests indicated strong capabilities, particularly in coding tasks. However, Google acknowledges that the model may occasionally generate incorrect information or experience slowdowns. Free access is currently provided, although it has not been integrated into the consumer Gemini app yet.

Source 🔗