Gemini 3.7 Flash Is Here:
Google's Newest Model, and the Pro Tier That Never Came

Published August 14, 2026 · Updated September 3, 2026

Update (September 3, 2026): Gemini 3.7 Flash has been superseded. Google released Gemini 3.8 Flash on September 2, 2026, and since September 3 it is what answers Gemini Web Search and Research on Secret Chat. The Pro tier described below is still missing. Everything else here describes the 3.7 Flash release as it stood in August and is kept for reference.

Google's Gemini 3.7 Flash is now live on Secret Chat AI. It answers your Gemini Web Search and Research questions as of today. Same private workspace where you already chat with GPT, Claude, Grok and Perplexity; a newer model behind two of the Gemini modes.

There's a second story in this release, and it's the more interesting one. Google's Flash line has shipped three times since May (3.5 in May, 3.6 in July, 3.7 in August) while the Pro line hasn't moved since January. Below is what we verified against Google's own API, what changed on Secret Chat, and what conspicuously didn't.

What Is Gemini 3.7 Flash?

Gemini 3.7 Flash is the newest model in Google's fast tier. We pulled its record straight from Google's public model catalogue rather than from a press release, and this is what it says:

  • Version 3.7-flash-08-2026 — an August 2026 build, live in Google's catalogue as we publish this.
  • 1,048,576-token context window with up to 65,536 tokens of output. That's a full million tokens of input: a large codebase, a contract set, or a long research thread in one conversation.
  • It is a thinking model. The catalogue marks it thinking: true, and it does reason before it replies; you can watch it deliberate on a hard question and answer a trivial one immediately.
  • Same tier, newer build. This is the fast tier getting sharper, not a new frontier model. If 3.6 Flash was already answering your questions well, 3.7 will feel like the same model on a slightly better day.

It Thinks Less, Not More

A thinking model spends time reasoning before it answers, and that time is the difference between a reply that arrives and a cursor that blinks. So rather than take the release note's word for anything, we measured it: the same prompts, sent to both models, comparing how much each one deliberated before speaking.

Gemini 3.7 Flash consistently reasoned less than 3.6 on identical prompts — 327 thinking tokens against 398 on one question, 193 against 205 on another. It reaches the same kind of answer with less deliberation, which you feel as a shorter wait.

That's worth measuring rather than assuming, because a newer model can just as easily go the other way (thinking harder and answering slower), and no spec sheet will tell you which one you're getting.

One thing to know about the price, since it's unusually good right now: Google launched 3.7 Flash at half the rate 3.6 Flash opened at, and its own documentation calls that an introductory price, running through 31 December 2026. It's a promotion with an end date, not a permanent repricing of the tier.

The Pro Tier That Never Shipped

Now the part nobody announced. Google's model line has two halves — Flash for speed, Pro for the hardest questions. Flash has moved three times since May. Pro hasn't moved since the start of the year.

Since we published this, the gap acquired an explanation — of a sort. Google launched 3.7 Flash on 13 August 2026 with no date for the Pro model, and Bloomberg and Axios both reported that Gemini 3.5 Pro, the flagship promised at May's I/O, remains delayed after DeepMind reportedly scrapped its original build over tool-calling failures. Treat that as reporting rather than as a Google statement: the company itself has still said nothing. The verifiable part is unchanged, and it's the part below.

There's no Gemini 3.5 Pro. There's no 3.6 Pro and no 3.7 Pro either. The newest Pro-tier text model in Google's public catalogue today is gemini-3.1-pro-preview, and its version string reads 3.1-pro-preview-01-2026: a January build, still carrying the word preview seven months later.

The cleanest evidence comes from Google itself. Google publishes moving aliases that always point at the newest model in each line, so you can simply ask the API what it currently considers latest. We did, and got this:

  • gemini-flash-latestgemini-3.7-flash
  • gemini-pro-latestgemini-3.1-pro-preview

Google's own pointer for "the latest Pro" still resolves to a preview build from January. That isn't a rumour or a reading of the tea leaves; it's what the API answers when you ask it.

We aren't going to speculate about why. Reporting has circulated suggesting the next Pro release was skipped and the effort folded into a larger future model, and that may well turn out to be right, but we haven't verified it and won't repeat it as fact. What is verifiable is the shape of the gap: three Flash releases between May and August, and a Pro tier still labelled preview since January.

What This Means on Secret Chat

It means our Gemini Research mode currently answers on Flash rather than on a Pro model, and we would rather say so plainly than quietly imply otherwise.

That slot used to run Google's Pro preview. When that build stopped being updated, we moved the slot to the newest Flash instead of leaving your hardest Gemini questions on a preview nobody had touched since January. If and when a genuine new Pro model ships, the slot moves back to it.

Here's exactly what runs where in the Gemini model today:

  • Smart Agentic (the default), a router reads your message and picks the handling. The router itself runs on Gemini 3.5 Flash Lite; where it sends you is what changed.
  • Web Search — now Gemini 3.7 Flash, with Google's own search grounding attached, so answers use current information rather than training-cutoff knowledge.
  • Research — now Gemini 3.7 Flash as well, at the deepest setting, for questions worth waiting on.
  • Fast Chat — unchanged, on Gemini 3.5 Flash Lite. It's the quickest tier in the family, and a full thinking model would be the wrong tool for a one-line question.
  • Image generation — unchanged. Gemini's image mode runs on Nano Banana, a separate image model that a text upgrade doesn't touch.

Or Do Not Bet on One Lab's Roadmap at All

The Pro gap is a good argument for the thing this app is actually built around. If a single lab can go seven months without shipping its top tier, then which lab you happened to pick is a worse question than what several of them say about the same problem. That's the AI Council: one question sent to several models from different companies at once, each answer revealed the moment that model finishes, and then a cheap referee model that extracts the factual claims into a disagreement table — which model asserted each claim, which contradicted it, which never mentioned it — plus a short synthesis of what at least two models agreed on and a list of things to verify before acting.

You pick the size: a Duet of two models, a Trio of three, or the Quartet, which runs ChatGPT, Claude, Gemini and Grok together. On the Duet and Trio you also choose which families take the seats. Every model on a panel comes from a different company, which is the whole point: two tiers of one family agreeing is far weaker evidence, and a lab having a quiet quarter matters much less when three others are answering the same question next to it. Panels seat each family's mid tier rather than its heaviest, because a council is evidence through disagreement rather than through any single answer being the deepest obtainable; depth from one model is what Research is for. When a question turns on current or checkable facts the whole panel answers with live web search attached, each provider using its own (Google's search grounding for the Gemini seat) decided once for the panel so the members stay comparable.

Two limits we would rather state than imply. Agreement is evidence, not proof: models share training data and can be wrong together, so disagreement is the signal worth reading. And a council widens where your question travels, since it reaches every model on the panel plus the referee. Each request is still anonymous, but anonymization protects who you are, not what you type. The finished check is stored only on your device, and the Session Privacy Report lists every model that took part.

How to Use Gemini 3.7 Flash on Secret Chat AI

Nothing to install or switch on — it's already live:

  1. Open the Secret Chat app and select the Gemini model in the model picker.
  2. Leave it on Smart Agentic and let the router decide, or pick a mode yourself next to the message box — Web Search or Research to get 3.7 Flash, Fast Chat for a quick reply.

A million tokens of context is the feature most people leave unused, so it's worth aiming a prompt at it deliberately:

I am going to paste a long document. Read all of it before you answer — don't skim and don't summarise it back to me. Then tell me the three things in here that would change a decision, quote the exact lines they come from, and flag anything that contradicts something else in the same document.

Private by Design, Whichever Gemini Answers

The upgrade changes the model, not the privacy. Secret Chat is a private gateway between you and the AI providers: your prompts are forwarded through our proxy, so you need no Google account and your requests are never tied to you on Google's side. We build no profile of you, and no chat is ever associated with your identity. Your history lives locally in your own browser, and a prompt exists on our side only for as long as it takes to fetch your answer; there's no stored chat archive on our servers. You can pay with crypto if you would rather not link a card to your AI use.

That matters with Gemini in particular. As we covered in Does Gemini Read Your Data?, a consumer Google account ties AI activity to the same identity as the rest of your Google life. Reaching Gemini 3.7 Flash through Secret Chat lets you use the same model as a stranger: retention may still apply at the provider, but your query reaches the LLM anonymized, not linked to your email or identity.

One honest caveat, as always. "Anonymously" describes the link, not the words: no account identifier travels with your prompt, but that isn't a claim that the text stops being identifying. Write your own name or your case number into a message and it's all still sitting there in the message. Secret Chat removes you from your queries (it doesn't remove the data from your messages), so redacting identifying details before you send is still your call. Anonymity is also not privilege, not a legal exemption, and not a way to put anything beyond the reach of a court that's entitled to it. For the full picture, see our dedicated Private Gemini page.

And if you would rather not bet on any single lab's roadmap, that's the whole idea behind having several. See why using GPT, Claude, Gemini, Grok and Perplexity together beats relying on one of them, and the AI Council described above is that idea built into the app.

Conclusion

Gemini 3.7 Flash is a real improvement in the tier where most questions actually get answered: a million tokens of context, and (measured, not assumed) slightly less deliberation to reach the same answer. It powers Gemini's Web Search and Research modes on Secret Chat as of today.

The Pro tier is the open question. Google's own "latest Pro" pointer still resolves to a January preview, seven months and three Flash releases later. When that changes, our Research mode moves with it.

Try it today at Secret Chat AI: the full multi-model lineup, your history in your own browser, and every query reaching the model anonymously.

Frequently Asked Questions

  1. What is Gemini 3.7 Flash?

    Gemini 3.7 Flash was the newest model in Google's fast Gemini tier when this article was published, released in August 2026 as version 3.7-flash-08-2026; Gemini 3.8 Flash superseded it on September 2, 2026. It's a thinking model with a 1,048,576-token context window and up to 65,536 tokens of output.

  2. How is Gemini 3.7 Flash different from 3.6 Flash?

    It's a newer build of the same fast tier rather than a new frontier model. In our own testing it reasoned slightly less than 3.6 before answering identical prompts, which you feel as a shorter wait. The context window is unchanged.

  3. Is there a Gemini 3.5 Pro?

    Not yet. Gemini 3.5 Pro was announced at Google I/O in May 2026 but hasn't shipped, and as of August 2026 there's no 3.5 Pro, 3.6 Pro or 3.7 Pro in Google's public model catalogue. The newest Pro-tier text model is gemini-3.1-pro-preview, a January 2026 build still labelled preview, and Google's own gemini-pro-latest alias still resolves to it while gemini-flash-latest resolves to Gemini 3.7 Flash.

  4. Why has Google shipped Flash three times but no new Pro?

    Google hasn't explained it. Bloomberg and Axios have reported that DeepMind scrapped the original 3.5 Pro build over tool-calling failures, which is reporting rather than a company statement, so we treat it as such. What is verifiable is the gap itself: the Flash line has moved through 3.5 (May), 3.6 (July), and 3.7 (August) while the Pro line has stayed on a January preview build.

  5. Which Gemini mode on Secret Chat uses 3.7 Flash?

    None any more. Web Search and Research ran on Gemini 3.7 Flash from August 14 to September 3, 2026, when Gemini 3.8 Flash replaced it in both modes; the Smart Agentic router still sends you to them automatically when your question calls for it. Fast Chat stays on Gemini 3.5 Flash Lite, and image generation runs on Nano Banana, a separate image model.

  6. Does Secret Chat's Research mode really run on Flash rather than Pro?

    Yes, and deliberately. That slot used to run Google's Pro preview; when that build stopped being updated we moved it to the newest Flash rather than leave it on a preview untouched since January. If a new Pro model ships, the slot moves back to it.

  7. What if I would rather not depend on Google's roadmap at all?

    Then use several labs on the same question instead of choosing one. The AI Council sends one question to several models from different companies at once — a Duet of two, a Trio of three, or a Quartet running ChatGPT, Claude, Gemini and Grok together — reveals each answer as that model finishes, and then has a referee build a disagreement table plus a short synthesis of what at least two models agreed on. Every model on a panel comes from a different company, since two tiers of one family agreeing is far weaker evidence. Agreement across them is still evidence rather than proof (models share training data and can be wrong together), so the disagreements are the part worth reading.

  8. Do I need a Google account to use Gemini 3.7 Flash?

    No. Secret Chat forwards your prompts through its own proxy, so you use Gemini without a Google account and without your requests being tied to you. No profile is built, no chat is associated with your identity, and your history stays in your browser. Retention may still apply at the provider, but your query reaches the LLM anonymized, not linked to your email or identity. Note that the content of your messages is sent verbatim: only your identity is removed.

  9. Can Gemini 3.7 Flash search the web on Secret Chat?

    Yes. The Web Search and Research modes attach Google's own search grounding, so answers combine the model's reasoning with current information instead of being limited to training-cutoff knowledge.