Google promised Gemini 3.5 Pro in June. It is late, has no price, and no benchmarks.
Five weeks after Sundar Pichai teased a 2M-token Pro model on the I/O stage, the rate card is still blank and the Vertex changelog still does not mention it. The one thing Google did make concrete is the part that touches your wallet: it reshuffled the subscription ladder, and its headline reasoning mode now starts at the $100 Ultra tier, not the entry plan.
Photo by Victoria Wang on Unsplash
Gemini 3.5 Pro was announced at Google I/O on May 19 with a 2M-token context window and a Deep Think reasoning mode. Google promised June. It is now late June and the model is still in limited Vertex preview, with no published API price and no benchmark numbers. Prediction markets have already shifted to a July launch. There is, however, one confirmed cost: at I/O Google rebuilt its consumer plans into a $20, $100, $200 ladder, and Deep Think, the generation's headline feature, sits on the Ultra rungs starting at $100, not the $20 entry plan. If you are budgeting around this model, ignore the spec rumors and plan against two real anchors instead: the current Gemini 3.1 Pro rate card and that subscription ladder. We walk through both below.
What Google actually put on stage
The I/O keynote gave Gemini 3.5 Pro about a minute. Pichai named it, said it would ship the following month, and pointed at two features: a 2M-token context window, twice what Gemini 3.5 Flash carries, and Deep Think, an extended reasoning mode that the company is positioning as the generation's headline. The Flash sibling actually launched that day. Pro got a promise.
What it did not get was substance. No model card, no system card, no pricing, and not a single benchmark score. That matters for a Pro tier, because the whole reason to reach for Pro over Flash is reasoning depth and context, and Google showed you neither a number nor a price for either. Everything specific you have read since, the exact token rate, a SWE-bench figure, a Humanity's Last Exam result with a 3.5 Pro label, was filled in by someone other than Google.
Why it has not shipped yet
The cleanest tell is the changelog. Google ships its model availability through the Vertex AI release notes, and as of June 25 there is no Gemini 3.5 Pro line in it. The most recent Pro entry is still Gemini 3.1 Pro going to preview back in February. A handful of enterprise accounts appear to have limited preview access, but a closed preview is not a launch, and it is certainly not an API you can build a product on this week.
The betting markets have already adjusted. Polymarket's Gemini-Pro release-date contracts now favor a July arrival over the original June window, with the money leaning toward a launch by the end of next month. Read plainly: the June promise is slipping, and July is now the consensus. None of that is alarming for a frontier model, late is normal, but it does mean anyone planning a migration around a June date should move that date.
The price Google did confirm
Here is the part nobody is talking about because it is not a token rate. At I/O, Google rebuilt its consumer subscriptions. The old single AI Ultra plan, which launched last year at $250 a month, is gone. In its place is a three-rung ladder that happens to match what OpenAI and Anthropic now charge almost to the dollar.
| Plan | Per month | What you get |
|---|---|---|
| Google AI Pro | $20 | Pro model access, roughly 4x free-tier limits. No Deep Think. |
| Google AI Ultra | $100 | 5x the Pro usage limits, 20TB storage, and Deep Think access. |
| Google AI Ultra (top) | $200 | 20x Pro limits, plus priority Deep Think and higher limits. |
The thing to notice is where Deep Think landed. The marquee capability of this generation, the reason a reasoning-heavy user would care about 3.5 Pro at all, starts at the $100 Ultra tier. The $20 plan gets you the Pro model but not its best mode, a 5x step up to reach the reasoning Google is building the launch around. That is a deliberate ladder: Google is using the new model to pull serious users up to the top rung, the exact same move OpenAI made with ChatGPT Pro and Anthropic made with Claude Max. We mapped that convergence when all three giants settled on the same $20/$100/$200 ladder, and 3.5 Pro is the bait at the top of Google's.
If you only use Gemini through the app, this is your real decision, and it is decidable today even though the model is not out. Deep Think at $100 versus the $20 Pro plan is a 5x jump for one reasoning mode. Most people do not need it. If you are running agents or long reasoning chains for work, you probably do, and you should budget the $100 now, or $200 if you want the priority access and higher limits, rather than be surprised by it at launch.
What the API will probably cost
For developers, the subscription ladder is beside the point; you pay per token. And there, the honest answer is that we do not know, so the useful move is to anchor on the model 3.5 Pro replaces. Gemini 3.1 Pro is live and priced like this:
| Gemini 3.1 Pro | Input / 1M | Output / 1M |
|---|---|---|
| Up to 200K tokens | $2.00 | $12.00 |
| Above 200K tokens | $4.00 | $18.00 |
Two things carry forward. First, the tiered cliff at 200K tokens: once a request crosses that line, the whole request reprices at the higher rate, not just the tokens past the threshold. A 2M-token window is a headline feature, but if 3.5 Pro keeps this structure, every long request you send lands in the doubled bracket. Plan your context budget around that, not around the marketing number.
Second, Google's recent direction has been to hold or only gently raise Pro pricing while pushing capability up, so a launch somewhere near the 3.1 Pro card is the reasonable base case. We are not going to print a precise 3.5 Pro number, because Google has not, and every specific figure floating around traces back to SEO guesswork rather than a Google page. The moment a real rate card appears, it goes on our pricing comparison page and into this post.
The field it is launching into
Whenever it lands, 3.5 Pro joins a frontier tier that has spread out a lot. These are the confirmed rates it will be measured against. Gemini 3.1 Pro is already the cheapest of the premium flagships per output token, which is the floor Google has to either hold or justify breaking.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| Grok 4.3 | $1.25 | $2.50 | 1M |
| Gemini 3.5 Flash | $1.50 | $9.00 | 1M |
| Gemini 3.1 Pro | $2.00 | $12.00 | 1M |
| Gemini 3.5 Pro (TBD) | not set | not set | 2M |
| Claude Opus 4.8 | $5.00 | $25.00 | 1M |
| GPT-5.5 | $5.00 | $30.00 | 1M |
| Claude Fable 5 | $10.00 | $50.00 | 1M |
| GPT-5.5 Pro | $30.00 | $180.00 | 1M |
Gemini 3.1 Pro reprices to $4 input / $18 output above 200K tokens. Claude Opus 4.8 shown at standard rate; its Fast mode is $10/$50. Claude Fable 5 has been suspended since June 12; its card is shown for reference. No official Gemini 3.5 Pro rate exists as of June 25, 2026.
The interesting tension is that Gemini 3.5 Flash, which already shipped, beat both Claude Opus 4.7 and GPT-5.5 on some agent benchmarks at $1.50/$9. That sets an awkward bar for Pro. If the Flash model is already winning agent tasks, Pro has to justify a higher rate with reasoning depth and that 2M window, not raw capability on everyday work. For a lot of production traffic, the cheaper Flash may stay the right call even after Pro arrives.
What to do while you wait
There is no migration to do this week, because there is nothing to migrate to. So treat the wait as prep time. If you use Gemini in the app, the only live decision is the subscription one, and Deep Think at $100 versus the $20 Pro plan is answerable right now from how much reasoning work you actually do.
If you build on the API, pull your real input and output token volume and run it through Gemini 3.1 Pro's numbers, including the 200K cliff if your prompts run long. That tells you the floor. When 3.5 Pro ships, you will be comparing against a figure you already understand instead of reacting to a launch post. Artificial Analysis will have independent quality-versus-price rankings the day it appears, which is the thing to trust over Google's own benchmark slide.
We will update this post the moment Google publishes a real model card and rate card. Until then, the cost calculator lets you price your workload on Gemini 3.1 Pro today, so you have a number ready before the new one drops.
Sources
- - Google I/O 2026 keynote coverage, May 19, 2026
- - Google Vertex AI generative AI release notes, read June 25, 2026
- - Google One AI subscriptions announcement, I/O 2026
- - Polymarket: next Gemini Pro model release-date market, read June 2026
- - Google Gemini API pricing page, Gemini 3.1 Pro and 3.5 Flash rates
- - TokenCost pricing data for current frontier models