Google is releasing Gemini 3.5 Pro today, after delaying the original launch to rebuild the model from scratch. It’s already generating the kind of “what is Gemini 3.5 Pro” search spike that follows every major AI release, so it’s worth a straight answer before you’re tempted to switch tools on the strength of a headline.
The standout spec is a 2-million-token context window — roughly enough to feed in an entire year of customer support conversations, a full contract library, or a large codebase in one go, alongside a new “Deep Think” mode for harder multi-step reasoning. For specialist use cases, that’s a genuine step change. For most UK SMEs, it’s a capability you’ll rarely touch.
What this actually changes for a small business
Context window size only matters when you’re regularly working with genuinely huge amounts of text in a single request — think a law firm reasoning across a multi-thousand-page case file, or a business ingesting a year of raw support tickets to find patterns. If your typical AI task is drafting emails, summarising a meeting, or answering questions about a product, you were never going to hit the ceiling of the previous generation of models, let alone this one.
The pricing reflects that positioning. Early reporting puts Gemini 3.5 Pro at a premium tier — well above the everyday models most SMEs already use inside Microsoft 365 Copilot, Google Workspace, or standalone tools like ChatGPT. Deep Think, the extended reasoning mode, is gated behind an even higher subscription tier. This is a model built for specialist, high-volume enterprise workloads, not a like-for-like upgrade for the tools your team uses today.
What’s genuinely worth paying attention to
That doesn’t mean ignore it entirely. If your business does have a large-document problem — due diligence, compliance archives, long-running case histories — a model that can reason across the whole thing at once, without chunking it into pieces first, can save real time. The question is whether that describes your actual workload, or whether it’s just a shinier version of a job you’re already doing fine.
This is where it’s worth getting outside input rather than guessing. ApplyAI works with UK SMEs specifically to match the AI tool to the actual task rather than the loudest launch announcement, and BuildApps can wire a model like this into your existing systems properly if there’s a real use case for it — rather than you paying premium-tier pricing for a capability that sits unused.
The pattern behind every launch like this
Every few months a new model arrives with a bigger context window, a longer benchmark table, and a wave of coverage asking whether you should switch. Underneath the headline number, the real question is always the same: what task, specifically, is slow, expensive, or error-prone right now, and does this new capability actually fix that? Most of the time the honest answer is that the model you already have access to inside your existing tools is more than capable — the bottleneck is how it’s being used, not which model is doing the work. Chasing every release adds subscription cost and integration effort without necessarily adding output.
The takeaway
New model launches will keep happening every few months, each with a bigger number attached — more tokens, more reasoning, more benchmarks. The businesses that get value from AI aren’t the ones chasing every release; they’re the ones who know what job they’re actually trying to do and pick the tool that fits it. Before you switch anything based on today’s news, ask what problem you’re solving. If you can’t name one that needs 2 million tokens of context, you don’t need to move yet.
