Video: "NEW Gemini 3.5 Pro LEAKS! Google Is Back and Will Rival Fable 5 & GPT-5.6!" by Julian Goldie on YouTube.
What the leaks actually say
The leaked details — which Julian Goldie covers with appropriate caution — suggest Google has been testing a model internally codenamed Cappuccino, positioned as the full Gemini 3.5 Pro release. The headline numbers are a 2-million-token context window (double what Flash currently offers publicly), a dedicated Deep Think reasoning mode for harder analytical tasks, and a background agent called Spark that runs across Gmail, Docs, Drive, Calendar and Search without requiring a prompt each time.
None of this is confirmed by Google. Leaks from large organisations are regularly accurate on direction but wrong on specifics — timing especially. Worth keeping that in mind before making decisions based on it.
The 2M context window: what it would actually change
A 2-million-token context window is genuinely large. To put it in concrete terms: you could feed the model your entire website, a competitor's entire website, six months of customer emails, and a full year of analytics data — and still have room left. That kind of capacity changes what's possible for SEO and content work. Not just "write me a blog post" but "here is everything published on this topic in the last year — now find the gaps and write something that genuinely adds to it."
In practice, very large context windows often come with a caveat: models can get less attentive toward the middle of an enormous input. Early builds of Cappuccino reportedly had a tendency to become less reliable on long-running tasks. Google will presumably address that before release, but it's worth watching for in testing.
Deep Think mode and the Spark agent
Deep Think is described as a reasoning mode that slows the model down deliberately to handle harder problems — similar in spirit to what OpenAI does with o-series reasoning or what Anthropic does with extended thinking. For SEO work, the practical use would be competitive analysis, strategy questions, or anything where you want the model to actually consider multiple angles rather than pattern-match to the most common answer.
Spark is the more interesting part from a search perspective. If it behaves as described — a persistent background agent running across Google's own apps — it would give Google a native personal AI with direct access to how you actually use Search, Docs and Gmail. That kind of ambient data is something no third-party tool can match. The SEO angle is straightforward: if Spark becomes how people interact with Google products day-to-day, the ranking signals it generates and responds to will matter. Nobody outside Google knows how this affects what gets cited or recommended yet.
What's genuinely useful to know right now
The honest answer is: not much has changed today. Gemini 3.5 Pro hasn't shipped. But the direction is clear — Google is building toward larger context, better reasoning, and a persistent agent layer integrated into its own ecosystem. That last point is the one with real SEO implications, and it's been true in different forms since AI Overviews launched.
What does change your strategy right now: content that answers specific questions clearly and factually, structured so an AI can pull the relevant passage without reading the whole page, backed by a site that Google already trusts. That approach works across whatever model Google deploys next, not just this one.
Where this connects to NordSys
We track how Google's AI changes what gets cited in Overviews and AI Mode answers — and adjust on-page structure, content depth, and authority signals accordingly. If you want your site positioned to benefit from whatever Gemini 3.5 Pro turns out to be when it ships, the groundwork is the same as it is now: clear answers, good structure, a trusted domain. We can help get that in place.
See our SEO & AI Ranking service →