|
TLDR
Gemini 3.5 Pro, Google's flagship, is announced and in limited preview but has not shipped. It is a reasoning model with multimodal understanding, not an image or video generator, and that distinction is the whole story for creators.
• Announced at Google I/O on May 19, expected this month, still in limited preview for business customers, with markets at even odds on a June launch.
• The expected specs: a two-million-token memory, a Deep Think reasoning mode, and the ability to read text, images, audio, and video together.
• The catch: it does not generate images or video. Google's creative generators are Imagen and Veo, which are separate.
• The use for creators: a thinking partner for the parts around generation, reading references, critiquing drafts, planning, and writing prompts.
|
|
•••
|
|
Google announced its flagship model, Gemini 3.5 Pro, at its developer conference in May, and then asked everyone to wait. It is coming this month, the company said, without naming a day. As of now it is still in limited preview for a handful of business customers, not yet in the app or available to the rest of us. The faster, lighter Gemini 3.5 Flash shipped that day instead, and is already live for everyone.
So this is a preview, not a review. But it is worth understanding before it lands, because the honest question for anyone making images, video, or music is not how good Gemini 3.5 Pro is. It is what a frontier reasoning model with eyes actually does for creative work. The answer is not generation. It is thinking.
Here is where it stands, what it will do, what it will not, and where it fits in a creative workflow once it arrives.
|
|
Where It Stands
Announced, previewed, not shipped.
|
|
Gemini 3.5 Pro was unveiled in mid-May alongside Gemini 3.5 Flash, but only Flash went live. On stage, Google's chief executive said the company looked forward to rolling Pro out the following month, a line that drew an audible groan from a crowd hoping for it that day. Since then it has been in internal use and a limited preview through Google's business cloud platform.
As of now, with the month nearly out, it still has not reached the public. Prediction markets put the odds of a launch by the end of June at roughly even. When it does ship, it will most likely appear first in Google's paid consumer plans and then for developers, announced through a single post with the full benchmark grid. Until then, treat any specific date you see as a guess.
|
|
What It Will Do
A bigger memory and a slower think.
|
|
Two capabilities define the Pro tier. The first is memory: a context window of around two million tokens, double Flash's and the largest of any production model announced so far. That is roughly a million and a half words held at once, enough to keep an entire project, brief, and reference set in a single session. The second is Deep Think, an extended reasoning mode that spends longer working through hard problems before answering, likely reserved for the higher-priced plan.
It is also natively multimodal, meaning it takes in text, images, audio, and video together and reasons across them in one request. It can look at your reference images and your draft and discuss them with you. On cost, expect roughly ten times Flash's rate, somewhere around fifteen dollars per million input tokens, available first through the twenty-dollar monthly plan, with Deep Think behind the two-hundred-fifty-dollar tier.
|
|
What It Is Not
Not a generator.
|
|
Here is the part that matters most for this audience, and the part the headlines blur. Gemini 3.5 Pro does not generate images or video. It is a reasoning and coding model with the ability to understand media, not to make it. When you read that it is multimodal, that means it can see and interpret a picture, not paint one.
Google's actual creative generators are separate products: its image models, Imagen and the one nicknamed Nano Banana, and its video model, Veo. Those are the tools you would prompt to make something. Gemini 3.5 Pro is the brain that sits beside them, not the hand that draws. Keeping that distinction clear is the difference between using it well and being let down that it does not do something it was never built to do.
|
|
What It Means for Your Work
A thinking partner, not a paintbrush.
|
|
Once you stop expecting it to generate, a reasoning model with eyes becomes genuinely useful in a creative process, just not at the moment of creation. Hand it a reference image and ask why it works, and it can read the light, the framing, the mood, and name what you were responding to. Show it a draft and it can critique the composition or flag what is pulling focus. It can help you plan a series, a shoot, or a sequence before you generate a single frame.
It is also a prompt collaborator. Describe what you want and it can help you write the prompt, or take one that is not landing and help you diagnose why. With a two-million-token memory, it can hold your whole project, your style notes, your past prompts, your references, and stay consistent across a long session. The generation still happens in your image and video tools. This is the layer that helps you think before and between.
|
|
The Takeaway
Watch for it, but know what it is.
|
|
When Gemini 3.5 Pro lands, and it should be soon, the useful response is not to expect a new way to make images. It is to gain a sharper way to think about them. The creators who get the most from frontier reasoning models are not prompting them for pictures. They are using them to read references, critique work, plan projects, and sharpen prompts, then taking that thinking back to the tools that actually generate.
So keep an eye out for the launch post, and in the meantime the lighter Flash is already live if you want to feel out the family. Just hold the distinction clearly: this is the brain, not the brush. Used that way, a model that cannot make a single image might still improve every one you make.
|
|
•••
I am putting together a critique pack: prompts for turning a reasoning model with vision into a creative collaborator, reading a reference, critiquing a draft, planning a series, and debugging a prompt that is not landing. Built for the thinking around generation, not the generation itself.
Want it when it ships? Reply with send me the critique pack and I will get it to you.
|
|
A QUESTION FOR YOU
What part of your process could use a second set of eyes?
Reply and tell me. The stage where you work alone, choosing references and judging drafts, is often the one a thinking tool helps most.
If this was useful, forward it to a creator who only thinks of these models as generators.
|
|
Until next time,
Luxe Prompting
|
|
Luxe Prompting
AI Image Generation for Creators
|
|
