Gemini

Google’s flagship AI assistant: multimodal conversations, Deep Think deep reasoning, and conversational video creation — deeply integrated across the entire Google ecosystem.

Gemini is Google’s flagship AI assistant, available on Web, iOS, Android, and — new as of April this year — a native Mac client. The model lineup is in the middle of a rapid refresh cycle: on July 21, Google shipped the mainline Gemini 3.6 Flash (Artificial Analysis measured 17% fewer output tokens than 3.5 Flash, at a lower price, with computer use now a built-in tool) along with the ultra-fast 3.5 Flash-Lite (350 token/s per Artificial Analysis), and announced a limited CodeMender pilot program for the security-focused 3.5 Flash Cyber. Meanwhile, 3.1 Pro targets complex tasks, 3.5 Pro is in testing with partners, and pre-training for Gemini 4 has already begun. And for research-grade challenges, the Deep Think deep-reasoning mode is there as a safety net.

For everyday users, the free tier already includes the mainline model — more than enough for daily Q&A, document analysis, and image generation. Power users can upgrade to Google AI Pro, while developers and professional creators get the Ultra tier, which bundles Spark, a personal agent for macOS. One thing to keep in mind: after Google I/O in May, billing shifted from daily message caps to a compute-based system, and high-compute workloads like video and image generation will burn through your quota noticeably faster.

Where Gemini truly stands apart is its ecosystem: deep integration with Google Search, Workspace, and Chrome. The Omni model can generate video directly through natural conversation, interleave text and images, and create custom AI personas, while Live Translate delivers real-time interpretation across 70+ languages with tone preserved. Competitors currently find it very hard to replicate this full multimodal suite end to end.

Points to note: no direct access from mainland China — you’ll need your own network setup and a Google account; free-tier users may be routed to lighter models during peak hours; and under compute-based billing, quota consumption is no longer intuitive, so it’s best to get a feel for your usage pace before deciding whether to go Ultra.

Model Iteration Timeline

Date	Version	Highlights
2023-03	Bard	LaMDA-based chatbot opens for early access
2023-12	Gemini 1.0	Ultra / Pro / Nano tiers launched, replacing PaLM 2
2024-02	Gemini 1.5 Pro	1 million-token long context; Bard rebranded as Gemini the same month
2024-12	Gemini 2.0 Flash	Built for the agent era, with a multimodal real-time API
2025-03	Gemini 2.5 Pro	Thinking reasoning model, topped the LMArena leaderboard
2025-11	Gemini 3 Pro	Rolled out across the entire Google suite on day one
2026-02	Gemini 3.1 Pro	Focus on long-horizon agents and complex tasks
2026-07	Gemini 3.6 Flash / 3.5 Flash-Lite / 3.5 Flash Cyber	Two models publicly available; limited pilot announced for the security-focused Cyber; 3.5 Pro in testing, Gemini 4 pre-training underway

Source: Agentopia · For the site-wide agent catalog, see /llms.txt