Google · 2025 · 11 · 18 · Model · ~2 min read
Google launched Gemini 3
What's actually new
- Real lead, not a tie. Gemini 3 took the top spot on most independent leaderboards at launch — first time Google clearly led, not just kept up.
- Genuine video reasoning. You can hand it a long video and ask 'what happened in minute 23?' or 'find the moment the cat appears'. Earlier models could caption frames, not reason about timelines.
- Million-token context window kept. Same as Gemini 2.5 — Google still leads on long documents.
- Gemini 3 Flash followed in December — the cheap, fast variant for high-volume API use.
If you want more
Worth knowing
- 'Best on planet' lasted weeks. Anthropic's Sonnet 4.6 closed the gap by February 2026.
- Pricing climbed versus Gemini 2.5 Pro for the higher tiers, especially with thinking enabled.
Who should care
Anyone working with long videos professionally — journalists, lawyers, educators, filmmakers. Developers building with the Google AI Studio free tier. Researchers tracking when (and whether) Google's research advantage finally turned into product wins.
What to do about it
Try Gemini 3 in Google AI Studio for free. Throw it a long PDF or a meeting recording — those are the tasks where it stretches its lead. For shorter chat, the comparison with GPT-5 and Sonnet 4.5 depends on your task.
Honest take
Gemini 3 was the moment Google stopped catching up and started leading. The real story wasn't the leaderboard score — Anthropic and OpenAI quickly closed it. The real story was video reasoning. For the first time, an AI could watch a long clip and answer questions about what happened across time, not just frame-by-frame description. That capability quietly reshapes what AI is useful for in 2026 — and it's still mostly Google's lane.
Sources
Last verified · 2026 · 05 · 05 · Found a fact wrong? corrections@aguidetocloud.com