Google's Gemini 3.5 Pro Is About to Land — 2M Tokens and 'Deep Think'

Why is a model that hasn't even shipped dominating the conversation?

It's rare for a not-yet-released model to dominate industry chatter. Gemini 3.5 Pro is doing exactly that. Google unveiled it at I/O on May 19, and it's on the doorstep of a June general availability (GA). In his keynote, Sundar Pichai said "give us until next month" — meaning June, with no committed date. As of early June, the model is still pre-GA, sitting in internal use and a limited enterprise preview.

The specs are heavy. A 2-million-token context window, plus a reasoning mode called Deep Think. 2M tokens is one of the largest contexts among production frontier models as of May 2026, and double its sibling Gemini 3.5 Flash's 1M. Deep Think pushes the model to reason more deeply and at length on hard problems — and interestingly, it's gated to the $250/month "Ultra" subscription tier, not the $20 Pro plan.

Pricing tells the position too. It's discussed at roughly $15 per million input tokens and $60 per million output — about 10x Gemini 3.5 Flash. So Gemini 3.5 Pro positions itself as a premium flagship for the hardest tasks, not a cheap mainstream model. In the same week xAI shouts "98% cheaper" and Moonshot releases open weights, Google differentiates from the opposite end — the "biggest, deepest model."

The players — Google, the Gemini family, and Pichai's "agentic era"

The first protagonist is Google (and DeepMind). Once said to be "behind in the AI race," Google counterpunched fast with the Gemini series. At I/O 2026, Pichai outright declared the "agentic Gemini era," and said the Gemini app crossed 900 million monthly active users as of May 19. Laying AI on top of the vast distribution of Search, Android, and Workspace is Google's biggest weapon.

The second protagonist is the Gemini 3.5 family itself, split into Flash (fast, cheap, mainstream) and Pro (big, deep, premium). Flash shipped first and built a huge base of 900M users; now Pro layers on top. I/O also unveiled "Gemini Omni," an any-to-any model, and "Gemini Spark," an Ultra-exclusive agent. So Pro isn't a one-off model — it's the apex of a family strategy.

The third protagonist is Sundar Pichai himself. His keynote message was clear: Google is no longer a chaser but a force redesigning entire products for the AI era. Yet he didn't commit a GA date for Pro, only "give us until next month." That signals the most powerful model needs more stability and safety vetting — while also seeding market anticipation that "it's coming soon."

The substance — Gemini 3.5 Pro by the numbers

Item	Detail
Unveiled	May 19, 2026 (Google I/O)
GA target	June 2026 (no fixed date)
Context	2M tokens
Key feature	Deep Think reasoning mode
Deep Think access	$250/month Ultra tier
Pricing (discussed)	~$15 in / ~$60 out (per M tokens)
Sibling	Gemini 3.5 Flash (1M, ~1/10 the price)
Current status	Internal use + enterprise preview

The most eye-catching is "2M context." Two million tokens means you can drop several thick books, a giant codebase, or a vast pile of documents into the model at once and ask questions. Less hassle chunking long context, and higher consistency since the model sees the whole context at once. For work where "long context equals quality" — legal, research, large-scale code analysis — it makes a decisive difference.

Gating Deep Think to the $250 Ultra tier is strategic: using the strongest reasoning as the differentiator of the top paid tier. Regular users get fast answers; heavy users and professionals willing to pay more get deep reasoning. It separates user segments by price — a sign that AI is fragmenting from a "$20-a-month egalitarian service" into "tiered, use-case-differentiated services."

One caveat, though: as of early June, Pro hasn't shipped. There's always a gap between announcement and actual rollout, and as Pichai's "next month" hints, GA could slip or roll out in stages. Benchmark numbers and pricing may also adjust at launch. So treat current Gemini 3.5 Pro less as a "finished product" and more as a "very specific trailer for something about to arrive."

What's in it for whom

Google wants to hold the "top of the frontier." Layering "the biggest, deepest model" (Pro) on a massive 900M-user base (Flash) completes a picture that captures both mass reach and peak performance. While rivals rush toward efficiency and value, Google differentiates with "ultra-large context and deep reasoning only we can do." Search, Android, and Cloud as distribution channels are the sturdy backbone of this strategy.

Enterprise customers benefit directly from the 2M context. Dropping in sprawling internal documents, giant codebases, and long legal or research material for analysis gets far easier. That's why it released first as a Vertex AI enterprise preview — enterprise workloads feel the value of long context most acutely. Add Deep Think and you can expect differentiated quality in complex analysis and decision support.

Heavy users and professionals gain a new option in the $250 Ultra tier. Pricey-looking, but for roles where deep reasoning meaningfully lifts productivity (developers, researchers, analysts), it can be worth it. A new user segment — "people for whom $250/month on AI is rational" — is forming. Conversely, for regular users Flash and the free tier suffice, so price tiering actually gives each group the right option.

Historical echoes — the wins and flops of "announced vs. shipped"

"Dazzling I/O announcement, but delayed or underwhelming actual launch" isn't rare in Google's history. Past AI demos said "coming soon" and quietly slipped, or weren't as smooth at launch as in the demo. So the market takes Pichai's "give us until next month" with equal parts anticipation and doubt. Google's announcements are always flashy; the question is whether it ships "exactly as promised, when promised."

There's a successful landing, though: the Gemini Flash line. As a fast, cheap model it rapidly captured the mass market and built a 900M-user base — Google proving to itself that "the right model on a vast distribution network spreads explosively." The Pro strategy stacks a "premium layer" on that success, so the base itself is solid.

But the competitive landscape has shifted. "Big context" used to be a powerful differentiator, but 2M-class contexts are now being matched by several frontier models. How long "Pro's exclusive edge" lasts is uncertain. Impressive at announcement, but by actual GA a rival may have similar specs out. So the real differentiator will be decided more by "real-world quality and ecosystem integration" than by raw numbers.

How rivals counter-play — OpenAI, Anthropic, and the open camp

OpenAI counters with "product integration and user base." ChatGPT already has a huge user base and a powerful brand, so it won't wobble from spec races over context size or reasoning modes alone. OpenAI is also pushing its own reasoning models and long context, leaning on being "the most familiar AI interface." Google's 900M vs. OpenAI's base — this distribution fight is the core battleground.

Anthropic differentiates on "coding and agentic reliability." Claude has built a strong reputation among developers for reliability in coding and agent work. When Google touts "the biggest context," Anthropic counters with "the most trustworthy task execution." Long context doesn't guarantee good results — putting in lots of context and accurately using that context are different abilities.

The open camp (Moonshot, DeepSeek) and the value camp (xAI Grok 4 Fast) press from the opposite direction. While Google aims up with a "$250 premium," they seize the bottom with "cheap or outright free." If the market polarizes into "premium vs. value," Google's Pro must constantly prove "it's worth the price." The 900M base is sturdy, but most of that base is free and low-cost users — worth remembering.

So what actually changes — by who you are

If you're a developer or enterprise, 2M context and Deep Think become a new option for "long-context, complex-reasoning" work. For workloads handling giant codebases or vast documents, it's worth testing directly on Vertex AI after GA. But since the price is 10x Flash, it's wise to use it selectively for work that truly needs long context and deep reasoning.

If you're in product or strategy, read the trend of "AI service tiering." Google gating Deep Think at $250/month signals AI moving from a "flat single price" to "use-case-differentiated pricing." This case can inform how you tier and price AI features in your own product.

If you're a general user, the free and low-cost tier (Flash) is plenty for now. Pro and Deep Think are for heavy users who want deeper performance for more money. Still, it's a "coming soon" trailer, so it's worth watching how it rolls out to regular users if it ships during June. Keep in mind there's always a gap between announcement and actual launch.

🥄 Three Things You're Probably Wondering

— So can I use it right now? Not yet. As of early June, Gemini 3.5 Pro is in internal use and enterprise preview, with general GA expected during June. Pichai only said "give us until next month," with no fixed date. Read it as a "coming soon" trailer.

— What's so good about 2 million tokens? You can drop several books' worth of material into the model at once and ask questions. Handling long documents and giant codebases whole, without chunking, raises consistency. But "putting in a lot" and "using it accurately" are different problems — real-world quality can't be called until after GA.

— $250/month for Deep Think — is it worth it? Depends on the person. It can pay off for professionals where deep reasoning lifts productivity (dev, research, analysis), but it's overkill for general use. Too early to say — only after launch, once Deep Think proves how much difference it makes on hard tasks, can you really judge the value.

Sources

• Google Gemini 3.5 Pro Nears June Launch With 2M Token Context And Deep Think — TechTimes
• Sundar Pichai's opening keynote — Google I/O 2026 (blog.google)
• Gemini 3.5 Pro: 2M Tokens, Deep Think Coming Soon — Enterprise DNA
• 100 things we announced at Google I/O 2026 — blog.google
• Gemini 3.5 Pro: The June 2026 Launch Guide — Codersera

Numbers are as of announcement and may change.