Explore by topic

The platform itself — Pro / Flash / Flash-Lite, the unified SDK, thinking, prompting, and the long-context window. Everything that sits under the headline name.

How modern Gemini reasons before answering. The thinking_level knob, the cost/latency story, and the 3.5 default-drop gotcha.

Image generation and editing across the Nano Banana family — Pro, 2, and Flash with search-grounded generation — plus Imagen 4 for its distinct photographic aesthetic.

Video generation with native synchronized audio. Veo 3.1 + Veo 3.1 Lite, image-to-video, camera direction, creative editing.

Audio side of the platform: Lyria 3 for music, Gemini 3.1 Flash TTS for steerable speech, and the real-time voice Live API.

Any-to-any: text, image, audio, and video in; video out with native synchronized audio. The model that collapses pipelines.

Multimodal embeddings — text, images, and audio all in the same vector space. The unified-search default.

Embodied reasoning for robots — spatial understanding, pointing, multi-step physical task planning, and the Robotics SDK.

The model that operates a browser: screenshot, decide, click, observe. The bridge to systems without APIs.

Autonomous research agents that plan, get reviewed, then execute for minutes to hours. Standard and Max tiers.

Google's first-party agent harness — the think-act-observe loop behind a clean async API. The single-agent default.

Isolated Linux environments for agents that need real code execution, package installs, and a persistent session filesystem.

Multi-agent framework. Python / TypeScript / Go / Java. Code-first agents that scale to production.

Runtime for ADK agents on Cloud Run or GKE, with Cloud Logging / Trace / Monitoring out of the box.

Two-tier memory: Sessions for within-conversation state, Memory Bank for durable cross-session memory (GA).

Model Context Protocol — vendor-neutral tools that any MCP-aware client can discover and use.

Using Gemini with LangGraph, LlamaIndex, CrewAI, and the Vercel AI SDK — when to mix, when to consolidate.

Identity, governance, and discovery for an agent fleet. DNS + IAM for the agent web.

Agent2Agent — the wire protocol agents use to call each other across vendors. Signed agent cards, verifiable identity.

Agent-led transactions with guard rails — per-transaction limits, allowlists, confirmation thresholds, audit logs.