Google Launches Gemini 3.5 Flash, Gemini 3.1 Pro, and Veo 3.1

Google has rolled out a major wave of updates across its artificial intelligence stack. Leading the developer release are **Gemini 3.5 Flash** and **Gemini 3.1 Pro**, pairing massive context sizes with low-latency agentic automation. Alongside the core text and multimodal models, Google also launched **Veo 3.1**, its next-generation generative video model designed for filmmakers and creative agencies.

Gemini 3.5 Flash: The High-Speed Multimodal Flagship

Gemini 3.5 Flash serves as the primary speed-oriented model in the Gemini API. Featuring a **1-million token context window**, 3.5 Flash processes audio, video, code, and text inputs simultaneously. The model is specifically optimized for high-frequency agent actions and "vibe-coding" workflows, where developers require sub-second latency feedback loops.

Pricing remains highly competitive, sitting at just **$0.075 per million input tokens** and **$0.30 per million output tokens**, presenting a direct challenge to OpenAI's GPT-5-mini.

Gemini 3.1 Pro: Deep Agentic Thinking

For complex tasks like multi-file software engineering, mathematical logic, and tool orchestration, **Gemini 3.1 Pro** is now the primary choice. With an expanded **2-million token context window**, 3.1 Pro features deep reasoning chains that run natively in the background, allowing it to navigate large codebases and documents with perfect retrieval accuracy.

Google also introduced **Gemini 3.1 Flash-Lite**, a micro-distilled model optimized for high-volume classification and simple formatting tasks, priced at just $0.03 per million inputs.

Veo 3.1: Cinematic Video Generation

Expanding its creative AI capabilities, Google launched the **Veo 3.1** video API. Delivering cinematic-quality 1080p generation with high temporal consistency and physics understanding, Veo 3.1 allows creators to generate detailed sequences using text prompts or reference images.

Veo 3.1 includes advanced camera control parameters (such as panning, zooming, and tracking shots) and embeds SynthID digital watermarks directly into the video frames to verify its AI origins.