Gemini 3 for developers: New reasoning, agentic capabilities (blog.google)

Google today unveiled Gemini 3 Pro for developers — a major upgrade aimed at agentic coding, multimodal reasoning, and single‑prompt “vibe coding.” The model arrives in preview via the Gemini API in Google AI Studio and Vertex AI (pricing: $2 per million input tokens and $12 per million output tokens for prompts ≤200k tokens; free-tier access in AI Studio with rate limits). Technically, Gemini 3 Pro sets new highs across benchmarks (top marks on MMMU‑Pro and Video MMMU for complex image/video reasoning, 1M‑token context window, WebDev Arena 1487 Elo, and 54.2% on Terminal‑Bench 2.0), and includes API features for a new “thinking” level, granular media resolution, and stricter thought‑signature validation to preserve multi‑turn reasoning. The release focuses on agentic workflows: a client‑side bash tool plus hosted server bash tool for safe shell proposals, richer grounding (Google Search + URL context) with structured outputs, and Google Antigravity — a public preview agentic IDE that coordinates intelligent agents across editor, terminal, and browser. Practical implications include faster end‑to‑end app generation from single prompts, deeper document and spatial reasoning for robotics/XR and screen agents, and high‑frame‑rate long‑context video understanding. For developers this means easier automation of coding tasks, more reliable multi‑agent architectures, and new multimodal applications that combine vision, long context, and tool use in production workflows.