Google unveils Gemini 3 AI model and AI-first IDE called Antigravity (arstechnica.com)

Google today unveiled Gemini 3 Pro — a limited-release, multimodal upgrade to its flagship LLM — and launched Antigravity, an AI-first integrated development environment. Google positions Gemini 3 as a step toward AGI, with expanded simulated reasoning and better understanding of text, images and video, plus more immersive visual outputs and reduced hallucinations. The company also touts “vibe coding” improvements that make the model more useful for developer workflows, and says Gemini 3 is being folded into its broader product stack. The release is notable for measurable gains on several public benchmarks: an LMArena ELO of 1,501 (≈50 points above Gemini 2.5 Pro), 72.1% on the 1,000-question SimpleQA Verified test, 37.5% on the PhD-level Humanity’s Last Exam without tool use, 23.4% on MathArena Apex, 1,487 ELO on WebDev Arena, and 76.2% on SWE-bench Verified for code generation. Those numbers signal progress in factuality, math, and coding performance but also underline remaining gaps (e.g., ~28% error on general knowledge). Antigravity pairs this model-level advancement with an IDE designed around AI-first coding, suggesting tighter model-tooling integration that could accelerate developer productivity and reshape how code is authored and reviewed — while reinforcing the need for human oversight given residual errors.