Google Launches Gemini 3 Pro at $2/Million Input Tokens

Google has released Gemini 3 Pro in preview at $2 per million input tokens and $12 per million output tokens for prompts of 200,000 tokens or less, available through the Gemini API in Google AI Studio and Vertex AI. The company describes it as "our most intelligent model that can help bring any idea to life" and says it "delivers unparalleled results across every major AI benchmark compared to previous versions." It is also free of charge, with rate limits, in Google AI Studio.

The release matters because it lands Google's flagship model directly in existing production agent and coding workflows, the segment where OpenAI, Anthropic and Google are competing hardest for developer mindshare and enterprise spending.

Agentic coding as the core pitch

Google frames Gemini 3 Pro as "a new foundation of intelligence for what's possible with an agentic coding model." The model scores 54.2% on Terminal-Bench 2.0, which tests a model's ability to use tools to operate a computer via terminal. Google also claims it surpasses Gemini 2.5 Pro at coding, "mastering both agentic workflows and complex zero-shot tasks."

On the WebDev Arena leaderboard, Gemini 3 Pro sits at the top with 1487 Elo.

The model is available across the developer ecosystem: Google's Gemini CLI, Android Studio, and third-party tools including Cursor, GitHub, JetBrains, Manus and Cline.

Google Antigravity

Alongside the model, Google introduced Google Antigravity, an agentic development platform built to showcase Gemini 3's capabilities. In Antigravity, "you act as the architect, collaborating with intelligent agents that operate autonomously across the editor, terminal, and browser." Those agents "plan and execute complex software tasks, communicating their work with the user via detailed artifacts," covering feature building, UI iteration, bug fixing, and research and report generation.

The public preview is free to download now for macOS, Windows and Linux from the Google Antigravity website.

New API tooling

Gemini 3 ships with a client-side bash tool that lets the model propose shell commands as part of agentic workflows — navigating local filesystems, driving development processes and automating system operations. Google pairs it with a hosted server-side bash tool for multi-language code generation and secure prototyping. Both are available in the Gemini API for early access partners, with general availability "coming soon."

The hosted tools Grounding with Google Search and URL context can now be combined with structured outputs. Google calls this "especially powerful for building agentic use cases which involve fetching and extracting data and then outputting them in a specific format for downstream agentic tasks."

The API update also adds a new thinking level, more granular media resolution parameters, and stricter validation for thought signatures, which Google says "is critical for preserving the model's thoughts across multi-turn conversations."

Multimodal and spatial reasoning

Google claims Gemini 3 is "the best model in the world for complex multimodal understanding," setting new highs on MMMU-Pro for complex image reasoning and Video MMMU for video understanding. It carries a 1 million-token context window. Developers can now configure multimodal vision processing with more granularity to trade off latency and cost against visual fidelity.

On documents, Gemini 3 Pro goes "beyond simple OCR (Object Character Recognition) to intelligently handle complex document understanding and reasoning," according to Google.

Its improved spatial understanding drives performance in embodied reasoning tasks — pointing, trajectory prediction and task progression — which Google says unlocks use cases in autonomous vehicles, XR devices and robotics. The same spatial reasoning powers screen understanding of desktop, mobile and OS screens, delivering "significant performance improvement for computer use agents." The model also infers user intent from mouse movements and screen annotations, demonstrated in a Visual Computer demo app.

For video, Gemini 3 Pro handles high-frame-rate understanding of rapid action and long-context recall that can "synthesiz[e] narratives and pinpoint specific details across hours of continuous footage."

'Vibe coding' in Google AI Studio

Google positions Gemini 3 Pro as unlocking "the true potential of 'vibe coding', where natural language is the only syntax you need." By improving complex instruction following and deep tool use, the model can translate a high-level idea into a fully interactive app from a single prompt, handling "the heavy lifting of multi-step planning and coding details."

Google AI Studio's Build mode now automatically wires up models and APIs for AI-native apps, with annotations for fast iteration. An "I'm feeling lucky" button lets Gemini 3 Pro handle "the creative spark and the code implementation simultaneously."

Google frames the release in market-shift terms: "The software landscape is shifting. As AI changes who builds and how they build, we are committed to meeting you where you are." The company calls this "just the start of the Gemini 3 era," with the Antigravity bash tools and broader agent capabilities pointing toward a wider rollout beyond preview pricing and rate limits.