Gemini 3: Google DeepMind’s most intelligent Artificial Intelligence model

Gemini 3 is Google DeepMind's most intelligent Artificial Intelligence model, combining advanced reasoning, native multimodality, and long-context understanding to help people learn, build, and plan. It is available via Gemini, Google AI Studio, the Gemini API, and integrations with developer platforms.

Gemini 3 is presented by Google DeepMind as the organisation’s most intelligent Artificial Intelligence model to date. The release highlights native multimodality, a massive context window, and improved reasoning and tool use so the model can follow complex instructions, call tools reliably, and handle simultaneous multi-step tasks. A specialised variant, Gemini 3 Deep Think, is described as a step-change for problems requiring creativity, strategic planning, and iterative improvement.

The Gemini family is surfaced as a suite for different needs, including 3 Pro for complex tasks, 2.5 Flash for fast everyday performance, and 2.5 Flash-Lite for high-volume cost-efficient use. Google positions Gemini 3 for developers and enterprises through multiple platforms and tools: Gemini itself, Google AI Studio, the Gemini API, Vertex AI Studio, and a new agentic developer experience called Google Antigravity. The site lists hands-on demos and example prompts-voxel art, procedural fractal worlds, retro videogames and more-and quotes partners and customers including Box, Cursor, Figma, GitHub, JetBrains, Rakuten, Wayfair, Thomson Reuters, Shopify, Manus AI and others describing improvements in multimodal understanding, frontend quality, coding accuracy, and business workflows.

DeepMind publishes a range of benchmark results on the page to quantify performance. Selected numbers for Gemini 3 Pro include 37.5% on Humanity’s Last Exam without tools (45.8% with search and code execution), 31.1% on ARC-AGI-2 visual reasoning, 91.9% on GPQA Diamond scientific knowledge, 95.0% on AIME 2025 mathematics without tools (100.0% with code execution), LiveCodeBench Elo 2,439 for competitive coding, and 77.0% on long-context MRCR v2 (128k). Multimodal measures such as MMMU-Pro (81.0%) and screen understanding (ScreenSpot-Pro 72.7%) are highlighted alongside agentic benchmarks like Terminal-Bench (54.2%) and τ2-bench agentic tool use (85.4%). The page also emphasises safety and responsibility, linking to evaluation methodology and safety reporting for Gemini 3 Pro as part of the model rollout.

72

Impact Score

Microsoft Artificial Intelligence launches seven new MAI models

Microsoft Artificial Intelligence announced seven new in-house MAI models spanning reasoning, coding, image, transcription, and voice. The company is pairing the launch with Frontier Tuning, a workflow-specific adaptation approach for enterprise-controlled models.

Artificial Intelligence designed vaccine targets coronavirus threats

University of Cambridge researchers have trialled a vaccine component designed entirely by Artificial Intelligence to train the immune system against a broad family of coronaviruses. Early safety work showed modest immune effects, while larger studies and related vaccine projects are underway.

Meta Instagram breach exposes Artificial Intelligence agent security gaps

Attackers exploited Meta’s Artificial Intelligence customer support agent to take over Instagram accounts, underscoring risks that go beyond advanced hacking models. Security researchers warn that agentic systems can create serious vulnerabilities when deployed without strong guardrails and red-teaming.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.