Gemini 3 Pro

2025-11-18 · Publisher : Google · Domain : Artificial Intelligence

Google launches Gemini 3 Pro, its most intelligent model: first on LMArena (1,501 Elo), strong in reasoning, multimodality and agentic coding, in preview in the Gemini API, Vertex AI and the app.

Google launches Gemini 3 Pro, its most intelligent model: first on LMArena (1,501 Elo), strong in reasoning, multimodality and agentic coding, in preview in the Gemini API, Vertex AI and the app. ### What changes - New Gemini 3 generation: PhD-level reasoning and state-of-the-art multimodal understanding. - Google's best-performing agentic coding and “vibe coding” model, with the Antigravity platform launched on this occasion. - New API controls: thinking level, media resolution, thought signatures. - Launched the same day in Google Search (AI Mode), the Gemini app, AI Studio, Vertex AI, Gemini CLI and at third parties (Cursor, GitHub, JetBrains, Replit). - Deep Think mode announced for later for Ultra subscribers. ### Specifications - Context: 1 million tokens input, 64,000 output; knowledge cutoff: January 2025. - Identifier: `gemini-3-pro-preview`. - Multimodal input (text, image, video, audio, PDF). ### Pricing | Tier | Input (USD / 1M tokens) | Output (USD / 1M tokens) | |---|---|---| | Prompts up to 200,000 tokens | 2 | 12 | | Prompts over 200,000 tokens | 4 | 18 | Free with rate limits in Google AI Studio. ### Benchmarks | Benchmark | Score | |---|---| | Humanity's Last Exam (without tools) | 37.5% | | GPQA Diamond | 91.9% | | MathArena Apex | 23.4% | | MMMU-Pro | 81% | | Video-MMMU | 87.6% | | SimpleQA Verified | 72.1% | | SWE-bench Verified | 76.2% | | Terminal-Bench 2.0 | 54.2% | | LMArena (Elo) / WebDev Arena (Elo) | 1,501 / 1,487 | Scores published by the vendor. ### Key takeaways Google takes the lead in several reasoning and coding rankings at the end of 2025. The model is offered in preview with an output price of 12 USD per million tokens.

Tags : LLM, Google, Gemini, Pro, Agentique