Google readies Gemini 3.5 Pro with a 2M-token context

Tired of an AI forgetting the start of a conversation as soon as it hits the third page of text? That is exactly what Google is trying to fix with Gemini 3.5 Pro. Unveiled at the I/O conference on May 19, the model is now in its final stretch before a general release expected in June.

The headline feature is a context window of two million tokens. In practice that means the model can hold an enormous amount of text in mind at once, enough for several thick books or an entire company’s document base. The previous Gemini 2.5 Pro worked with one million tokens, so this doubles the capacity. For anyone feeding in long contracts, transcripts or code, that is the difference between the model seeing the whole picture or just a slice of it.

The second new feature concerns the Deep Think mode. It is a way of working where the model does not rush to answer, but takes more time to reason through a problem step by step. The Pro plan gains basic access to this kind of reasoning, but the full Deep Think mode remains tied to the pricier Ultra subscription at 250 dollars a month, while Pro costs 20 dollars a month. The practical payoff is better handling of tasks that demand reasoning, such as harder code, mathematics or multi-step analysis.

What does an ordinary user get, and what does a developer get?

It helps to split this into two groups. If you use Gemini through a phone app or through search, the biggest change you will notice is that the model holds context better and loses the thread less often in long conversations. Basic Deep Think will likely be offered as an option when you need a more serious answer, at the cost of a slightly longer wait, while the full mode stays behind the Ultra subscription.

For developers and companies the story is more concrete, because it all comes down to cost. According to available estimates, Gemini 3.5 Pro could cost around 15 dollars per million input tokens and around 60 dollars per million output tokens. That is roughly ten times more than the cheaper Flash model. The logic is clear: Flash for high-volume, simple tasks, and Pro for the hard work where paying more pays off. Our advice is to use Pro selectively, where reasoning quality genuinely changes the outcome, rather than for every query.

It is worth keeping in mind that these are announced figures and expectations, not something we have tried in everyday use. Prices and exact limits often get tuned only once a model ships to everyone.

Why is Google pushing the context limit right now?

In recent months the AI race has been less about who has the smartest model and more about who can process the most information at once. A large context is exactly that. Companies find it appealing to drop their entire internal documentation into a single prompt instead of chopping it into pieces and tracking what sits where.

Gemini 3.5 Pro pushes the large context and basic Deep Think down into a more affordable 20-dollar-a-month plan here, so Google effectively offers serious capacity without the priciest tier. Full Deep Think and the Ultra tier (250 dollars a month) still exist as the premium option for anyone who needs the strongest reasoning. For users that could mean a clearer ladder of plans, but only once we see real performance and final pricing will we know whether Pro has become the default pick or a luxury for special occasions.

Would a larger context window actually change the way you use AI, or do you care more about an answer arriving fast and cheap?