Gemini 3.7 Flash Is Live: Half the Price, 1M Context, and What's Actually Confirmed

Gemini 3.7 Flash is live, and Google's own CEO is now on the record about it. The model shipped on August 13, 2026 — just three weeks after Gemini 3.6 Flash — and Sundar Pichai has since confirmed the launch publicly, saying Gemini 3.7 Flash "smashed previous Gemini growth records in its first week" and is the fastest-growing Gemini model Google has shipped. That growth figure is executive-stated; Google has not published usage numbers. The model is already running inside Google Search's AI Mode and the Gemini app, not just behind an API key. And the pricing held to the leak: $0.75 per million input tokens and $3.75 per million output, half of Gemini 3.6 Flash's $1.50/$7.50.

Gemini 3.7 Flash is a multimodal model Google's model card describes as its latest and most capable Flash model, built for complex coding, agentic workflows, and reliable multi-step execution. Google's launch framing leads with gains in coding, knowledge work, and web development — vendor claims, examined below — while the specs are already on the record: the same 1 million-token context window the Flash tier has standardized on, a 64K-token output, and availability through the Gemini API (including Google AI Studio), Google Antigravity, Gemini Enterprise, and Gemini Spark for AI Pro and Ultra subscribers. Since August 14 the model has also been running inside Google Search's AI Mode and the Gemini app, where it replaced Gemini 3.6 Flash in the model picker. What the model is, what it costs, and where it runs are now a matter of record; the open question is what it actually delivers.

One piece of discipline from the leak coverage carries straight into a launch piece: Google's performance claims are vendor-reported until an independent measurement confirms them. The first independent number arrived within a day — Artificial Analysis scores Gemini 3.7 Flash at 56 on its Intelligence Index at high reasoning, four points above the 52 Gemini 3.6 Flash scored at launch and, per AA, on the Pareto frontier of intelligence versus speed. Google's own benchmark figures for coding, knowledge work, and web development remain the vendor's word. The same discipline covers the newest pitch: on August 19, Google's developer account demoed Gemini 3.7 Flash building an animated parallax landing page from a single prompt by orchestrating Nano Banana and Omni, Google's image- and video-generation models — impressive as a demonstration, and still Google's word until someone outside the company reproduces it. Treat the price, the context window, and the availability as facts. Treat the performance deltas and the demo as promises under test.

From leak to launch: the signals that checked out

Because this model was first tracked here as an unverified leak, it is worth recording how the leak aged — it is the difference between a rumor and a pattern the Flash tier has now demonstrated repeatedly.

• The August 7 X post from @chetaslua said groundwork for Gemini 3.7 Flash was "in place on their side," with no date. Six days later the model shipped. The "sooner rather than later" read was right.

• The August 7-8 references to gemini-3.7-flash inside Google's own Python GenAI SDK on GitHub pointed at a real model ID — the same ID now present in the Gemini API.

• The August 13 report that Gemini 3.7 Flash was staged on Gemini Enterprise in a hidden state with a possible same-day release was accurate about the date, and the rumored $0.75/$3.75 price was accurate about the price.

• The capstone came in the model's first week, when Sundar Pichai publicly confirmed the launch on X and called Gemini 3.7 Flash Google's fastest-growing Gemini model to date. An unverified August 7 X post resolved into a CEO-confirmed launch within a fortnight — the cleanest possible outcome for a leak.

None of that makes leak coverage redundant — these same signals could have resolved to "delayed" or "renamed." But a tier that ships this fast, this quietly, and this cheaply is exactly why Google keeps doing its most interesting work in Flash.

What's confirmed versus what's still a claim

Now that the model is live, the ledger from the leak post flips: most of what was "reported" has moved into the confirmed column, and the remaining uncertainty is about performance, not existence.

• Confirmed — Google has published a Gemini 3.7 Flash model card; the model is available through the Gemini API, Google AI Studio, Google Antigravity, Gemini Enterprise, and Gemini Spark; since August 14 it also runs inside Google Search's AI Mode and the Gemini app, where it replaced Gemini 3.6 Flash as the picker's Flash model; the announced price is $0.75/$3.75 per million tokens through December 31, 2026; the context window is 1M tokens with a 64K-token output.

• Executive-stated — Sundar Pichai says Gemini 3.7 Flash "smashed previous Gemini growth records in its first week" and is Google's fastest-growing Gemini model to date; Logan Kilpatrick, who leads Google's AI Studio, echoed the fastest-growing framing. No usage numbers have been released, so this is Google's claim about its own momentum, not an independently verifiable figure.

• Independently measured — Artificial Analysis put Gemini 3.7 Flash on its Intelligence Index within a day of launch: 56 at high reasoning (53 at medium, 51 at low), four points above the 52 Gemini 3.6 Flash scored at launch, with the gains AA attributes largely to agentic evaluations.

• Vendor-reported — Google's own eval figures for the headline use cases: 43.6% on FrontierCode 1.1 (up from 34.4%), 65.3% on DeepSWE v1.1, an Elo of 1588 on the web-development arena where Google says it ranks #1, 34.0% on GDP.pdf, and 30.4% on AutomationBench. These are Google's numbers until an outside party reproduces them.

• Vendor demo — On August 19, Google AI Developers published a demo of Gemini 3.7 Flash producing an animated parallax landing page from a single prompt by orchestrating Nano Banana and Omni, Google's image- and video-generation models. A demonstration is Google's word until an outside party reproduces it; the demo gets its own section below.

• Not yet known — whether Google extends the introductory rate past December 31, 2026 (it doubles to $1.50/$7.50 on January 1, 2027), whether the gains change the calculus that has kept Gemini 3.5 Pro unreleased, and how the model holds up on workloads the vendor did not advertise.

For context on those "gains over Gemini 3.6 Flash" claims, this is the independent baseline Google is being measured against:

Gemini 3.6 Flash's independently measured Artificial Analysis Intelligence Index of 52 (ranked #26 of 185 models at launch) and its $1.50/1M input price are the numbers Gemini 3.7 Flash has to beat — and the 3.7 price point already undercuts the 3.6 input price by half before any performance comparison begins.

The demo: one prompt to a parallax landing page

The freshest signal arrived after launch. On August 19, 2026, Google AI Developers published a demo that is the clearest picture yet of what Google built Gemini 3.7 Flash to do: from a single prompt, the model produces a fully animated parallax landing page — copy, imagery, and motion included — by deciding on its own which tools to call. Google shows Gemini 3.7 Flash orchestrating Nano Banana, its image-generation model, and Omni, its video-generation model: the Flash model writes the copy, calls Nano Banana for image references, and calls Omni to render parallax-ready video loops, then assembles the page. It is Google's own demonstration of its own model — vendor-stated, not independently verified.

A demo is a promise, not a benchmark. It establishes the direction Google is pushing the Flash tier — an orchestrator model that does not just answer but acts, calling specialized models as tools — but it leaves the questions that matter open. Whether the one-shot result is production-grade or demo-grade, whether the workflow reproduces outside Google's own tooling, and how much hand-tuning sat between the prompt and the published page are all things nobody outside Google has tested. What is on the record is positioning: Google is publicly casting its workhorse model as the conductor of its image and video models.

The pattern the demo shows — one model deciding which specialized model to call, then assembling their outputs — is the same shape that makes a routing layer useful. OrcaRouter's routing DSL composes several models into one call, model fusion runs a panel of models on the same task, and automatic failover routes around errors or quality regressions, so testing a multi-model orchestration like the one in the demo is a configuration change rather than an integration. If one-prompt website building matures from demo to product, that is the setup you would run it on: a slice of traffic to the new orchestration, the proven path as the safety net.

Why the price is the headline

The half-price is the most consequential part of this launch, and the arithmetic is simple. On a 1M-token input / 1M-token output workload, Gemini 3.6 Flash costs $1.50 plus $7.50, or $9.00. Gemini 3.7 Flash at the announced rate costs $0.75 plus $3.75, or $4.50 — a 50% cut on the same volume, and if Google's claimed efficiency improvements hold, the effective cost per completed task drops further still.

Two caveats worth writing down. First, the $0.75/$3.75 rate is introductory and time-boxed: Google has confirmed it runs through December 31, 2026, and doubles to $1.50/$7.50 on January 1, 2027. Second, some platforms are currently running promotional pricing below the announced rate, so check the live price at the point of call for what you will actually pay — the announced rate is the reliable floor, but it is a floor with a date on it.

The Gemini 3.5 Pro backdrop still matters

Google's flagship Gemini 3.5 Pro still has not shipped as of August 14, 2026 — months after I/O, with Bloomberg reporting in mid-July that it was behind schedule over coding capability and later reports describing internal deadline slippage. The Flash tier keeps delivering while the Pro tier stalls, which is why a half-price, faster Flash is a bigger event than a typical point release: it is the proof that Google's workhorse line is where the company is actually competing.

Don't confuse it with Qwen 3.7 Flash

Both "3.7 Flash" models now exist as shipped models, and they could not be more different.

• Gemini 3.7 Flash — Google's new Flash-tier model at $0.75/$3.75 per million tokens, 1M context, multimodal, announced August 13, 2026.

• Qwen 3.7 Flash — Alibaba's vision-language model that shipped quietly in late July 2026 at roughly $0.03/$0.13 per million tokens with a 1M context window, aimed at multimodal agents and computer-use tasks.

The price gap alone — roughly 25x on input tokens — separates them, and the vendor does too. When someone says "I'm running 3.7 Flash," the first question is still which one.

Should you move from Gemini 3.6 Flash?

The honest answer splits in two. For cost, the announced half-price is a real lever: if you pay meaningful volume on Gemini 3.6 Flash, the same workload on Gemini 3.7 Flash costs half as much on tokens, before any claimed efficiency gains. That is a switch worth modeling on your own call patterns. For trust, the picture improved within a day of launch — the independent Artificial Analysis Index is already in at 56 versus 3.6 Flash's 52 — but the specific coding, knowledge-work, and web-development deltas Google is selling are still vendor-reported, and Gemini 3.6 Flash remains the stable, documented default. The standard play is to keep a proven model as the default and route a slice of traffic to the new one, comparing quality on your own tasks.

Neither move requires a project if you are on a routing layer. Gemini 3.6 Flash and Gemini 3.5 Flash are both available today through a single OpenAI-compatible endpoint on OrcaRouter at provider list price with no markup — so a Google price cut anywhere in the tier shows up on our side the same day, and a later switch from one Flash model to another is a model-name change rather than an integration. Automatic failover is the pattern for evaluating a just-shipped model however it is served: point a slice of real traffic at it, let the router fall back to a proven model on errors, rate limits, or obvious quality regressions, and keep the old model as the safety net until the new one earns the traffic. The routing DSL and model fusion let you compose the new model with others in a single call, so testing it against your own workload is a configuration change.

What to watch next

• The next independent checks — Artificial Analysis has already published its Intelligence Index (56 at high reasoning), so the benchmark question has moved from "is the gain real" to whether Google's coding and web-development claims survive external reproduction. The August 19 demo adds a second, separate question: does one-prompt website building hold up outside Google's own tooling, and how much of the result was hand-assembled?

• The price — the $0.75/$3.75 rate is now a dated fact: it holds through December 31, 2026, then doubles. Watch whether Google extends it, and whether third-party promotional pricing normalizes to it.

• The rollout — Gemini 3.7 Flash in Google Search's AI Mode currently sits behind the Google AI Pro and Ultra subscription and is English-only; watch whether Google widens it to free tiers and promotes it to AI Mode's default, the pattern Flash models have followed before.

• Gemini 3.5 Pro — if Google's flagship finally ships, the "Flash is where Google delivers" framing of the past month may shift.

FAQ

Is Gemini 3.7 Flash released?

Yes — on August 13, 2026, and Google CEO Sundar Pichai has confirmed the launch publicly, calling Gemini 3.7 Flash Google's fastest-growing Gemini model to date. Google has published a Gemini 3.7 Flash model card describing it as its latest and most capable Flash model, the gemini-3.7-flash ID is served through the Gemini API, and Google lists it across Google AI Studio, Google Antigravity, Gemini Enterprise, and Gemini Spark, with the model now also live in Google Search's AI Mode and the Gemini app. This is the model whose preparation was reported here as a leak on August 7.

How much does Gemini 3.7 Flash cost?

$0.75 per million input tokens and $3.75 per million output at the announced introductory price — half of Gemini 3.6 Flash's $1.50/$7.50, per Google's announcement. The rate is confirmed through December 31, 2026, and doubles to $1.50/$7.50 on January 1, 2027. Some platforms currently run promotional pricing below the announced rate, so the live rate at the point of call is what you will pay.

Is Gemini 3.7 Flash in Google Search and the Gemini app?

Yes. The day after launch, on August 14, 2026, Google added Gemini 3.7 Flash to AI Mode in Google Search and replaced Gemini 3.6 Flash with it in the Gemini app's model picker across Android, iOS, web, and macOS. The Search integration currently requires a Google AI Pro or Ultra subscription and is English-only; in the Gemini app the model sits in the regular model picker, and it remains reachable through the Gemini API, Google AI Studio, Google Antigravity, Gemini Enterprise, and Gemini Spark.

Can Gemini 3.7 Flash really build a website from a single prompt?

In Google's own demonstration, yes — on August 19, Google AI Developers showed Gemini 3.7 Flash producing a fully animated parallax landing page from one prompt by calling Nano Banana for image references and Omni for parallax-ready video. It is a vendor demo in Google's own tooling, not an independent test: nobody outside Google has verified how production-ready the result is or how much tuning sat between the prompt and the page.

Should I upgrade from Gemini 3.6 Flash to Gemini 3.7 Flash?

The price is half and the first independent score is in — Artificial Analysis Index 56 at high reasoning, up from 3.6 Flash's 52 — but Google's headline coding, knowledge-work, and web-development deltas are still vendor-reported. Standard practice: keep Gemini 3.6 Flash as the default, evaluate Gemini 3.7 Flash on your own workload with failover, and switch when it earns the traffic.

Gemini 3.7 Flash is the Flash tier doing exactly what the Flash tier does — shipping fast, shipping cheaper, and this time arriving with an independent score that beats its predecessor. The leak was right about the name, the date, and the price; the CEO has now publicly confirmed the launch, and the remaining growth and performance claims are still under test — the "fastest-growing" figure with no published numbers, the coding and web-development deltas still vendor-reported, and Google's one-prompt website demo in the same vendor-claim column. If cost-per-task is your lever, the half-price makes Gemini 3.7 Flash worth benchmarking today. If stability is your priority, Gemini 3.6 Flash remains the proven choice — and moving later is a model-name change, not a project.