Product launch · Xiaohu explains

Google shipped three Gemini models at once, and none of them is the 3.5 Pro everyone was waiting for

All three sit on the Flash line: the workhorse writes two-thirds fewer words, the cheapest tier actually raised its text output price by two-thirds, and the third model goes only to governments and trusted partners.
The one-minute version
  • On July 21, Google shipped three Gemini models at once, all on the Flash line: the workhorse 3.6 Flash, the fastest and cheapest member of the 3.5 family, 3.5 Flash-Lite, and a bug-hunting specialist, 3.5 Flash Cyber.
  • The long-awaited 3.5 Pro never showed up. The flagship Pro line is still stuck at 3.1 Pro Preview, and the launch page offers a single line about "testing with partners" — while also announcing that pre-training for Gemini 4 has begun.
  • The pitch for 3.6 Flash is that it talks less. On the same batch of software engineering tasks, output dropped from 276K tokens to 97K, the success rate climbed from 37% to 49%, and the output price fell from $9.00 to $7.50.
  • 3.5 Flash-Lite pushes 350 output tokens per second, the fastest in the 3.5 family — but its output price is up 66.7% from the previous generation, and the launch page only calls it "compelling value."
  • 3.5 Flash Cyber isn't for sale. It scores 83.2% on CyberGym, fourth among the five systems in Google's own chart.
Figures here come from Google's own launch page and the DeepMind blog; benchmark numbers are Google's internal runs. Per a DeepMind blog footnote, competitor CyberGym scores are self-reported by each vendor. Prices and version status were checked against the official Gemini API pricing page.
What shipped

Three models in one drop, and the one everybody wanted isn't among them

On July 21, 2026, Google released three Gemini models at once. All three belong to Flash, the cheap-and-fast product line. Not a single flagship.

Workhorse · available today
Gemini 3.6 Flash
A generational update to the tier that does the everyday work. Coding, knowledge work, and multimodal all improve — but the headline is that it cuts down on word count.
Input $1.50Output $7.50API / AI Studio / Gemini app
Fastest in 3.5 · available today
Gemini 3.5 Flash-Lite
The low-latency, high-throughput tier. The launch page names agentic search and document processing as its target workloads.
Input $0.30Output $2.50350 tokens/sec
Specialist · not for sale
Gemini 3.5 Flash Cyber
Fine-tuned from 3.5 Flash for exactly one job: find software vulnerabilities, confirm them, and write the patch. Not publicly available.
Limited pilot comingGovernments and trusted partners onlyDelivered via CodeMender, Google's code security agent
Missing
Gemini 3.5 Pro
The launch page mentions it twice: near the top, that it's in testing with partners and will go wide when ready; at the end, that Google looks forward to shipping it soon. No date, no benchmarks, no price.
In partner testing

The newest thing on the Pro line is still 3.1 Pro Preview

Check the official Gemini API pricing page and the latest Pro model listed is gemini-3.1-pro-preview — still a preview. Both of today's releases, 3.6 Flash and 3.5 Flash-Lite, are generally available.

Flash line · workhorse tier
3 Flash Preview 3.5 Flash 3.6 Flash · today
Flash-Lite line · cheapest tier
2.5 Flash-Lite 3.1 Flash-Lite 3.5 Flash-Lite · today
Pro line · flagship tier
3.1 Pro Preview · still newest 3.5 Pro · missing
The newest public model on the Pro line remains 3.1 Pro Preview, still in preview. 3.5 Pro was not part of this release.
Next generation
Gemini 4 · pre-training started

The launch page closes with a note on what's next:

We have started our most ambitious pre-training run yet, for Gemini 4, and progress is exciting. Google Blog, 2026-07-21
Official demos

Eight Demo Lab clips showing the two available models at work

Alongside the main text, the launch page carries a set of Demo Lab videos showing what the two shipping models do inside real tools.

Four clips for 3.6 Flash

Parsing financial data and meeting transcripts
Using Managed Agents in AI Studio to parse and analyze financial data and transcripts more efficiently and accurately than 3.5 Flash.
Google AI Studio
Code migration, two models side by side
Running a code migration with multi-agent orchestration in Antigravity, at lower latency and higher quality than 3.5 Flash.
Google Antigravity
Building your own creative tool on the spot
Building a photo texture extractor for 3D workflows from scratch, inside canvas in the Gemini app.
Gemini app · canvas
From concept to interactive prototype
Building an interactive theme studio in Antigravity with the tldraw offline editor, leaning on the model's strong visual understanding.
Google Antigravity

Four clips for 3.5 Flash-Lite

Pulling product attributes from a huge e-commerce dataset
Extracting and synthesizing product attributes from a massive e-commerce dataset.
High-throughput batch
25 web designs in one shot
3.6 Flash acts as the lead agent handing out work while 3.5 Flash-Lite does the volume, instantly producing 25 distinct designs you can click into.
Two models together
Receipt scanning and translation at volume
Receipt analysis and translation at scale.
High-throughput batch
Standing up a playable game fast
Generating several gameplay concepts at once, then iterating round after round straight into a working game.
Rapid iteration
These eight clips were produced and published by Google as official launch material. The descriptions here follow the captions Google attached to each video; for the receipt analysis and translation clip, Google supplied only a title and no description.
The workhorse

3.6 Flash sells itself on brevity: two-thirds fewer words on the same work

On the same batch of long-horizon software engineering tasks, the previous 3.5 Flash averaged 276,000 tokens before it was done. 3.6 Flash spends 97,000. Nearly two-thirds less talk — and the share of tasks it actually gets right rises from 37% to 49%.

Both numbers come from Google's published charts, drawn from the same benchmark: DeepSWE v1.1. It measures the kind of software work a model has to grind through on its own — read a real codebase, locate the problem, change the code, run the tests, then keep going based on what comes back.

DeepSWE v1.1 · two metrics, one benchmark
Average output per task (shorter is cheaper)
3.5 Flash
276K
3.6 Flash
97K
Down 64.9% — a little over a third of what it was.
Pass rate on the same tasks (higher is better)
3.5 Flash
37%
3.6 Flash
49%
Up 12 percentage points.
On an OSWorld verified task run through the API, 3.6 Flash is more token-efficient and less verbose than 3.5 Flash. Source: original caption on the Google Blog video
Chart comparing average output tokens for 3.6 Flash
Left: average output on DeepSWE v1.1 falls from 276K to 97K tokens. Right: on the Artificial Analysis Intelligence Index set, from 28K to 23K. Source: Google Blog

Google quotes the reduction two different ways: 17% by the Artificial Analysis index, and "up to 65%" on "certain evals, such as Datacurve's DeepSWE." Its explanation of the mechanism is two lines long — fewer reasoning steps and tool calls to finish a multi-step task, plus higher precision, which means fewer redundant code edits and fewer execution loops.

Output costs five times what input does, so writing less is saving money

What you send the model counts as input; what the model sends back counts as output, and the two are billed separately. Here's the spread for 3.6 Flash on the Gemini API pricing page:

Input
$1.50
per million tokens
what you send it
Output
$7.50
per million tokens
what it sends back (5× the input price)

The pricing page marks the output column "including thinking tokens" — the model's own reasoning is billed as output too. And an agentic workflow is precisely the case where the model works through the steps itself: read a file, call a tool, look at the result, decide the next move. Stack up enough steps and that accumulated output is your bill.

One task, tokens piling up step by step Each block = one step the model takes on its own: think, write a plan, call a tool 3.5 Flash 276K More reasoning steps and more tool calls 3.6 Flash 97K Google reports fewer steps and fewer tool calls Block counts and widths are illustrative; they don't represent actual step counts.

Price: output down 16.7%, plus two tiers the launch page skipped

The launch page says "lower cost per output token" and lists 3.6 Flash at $1.50 in, $7.50 out. Cross-checking the Gemini API pricing page: 3.5 Flash also charges $1.50 for input, but $9.00 for output — so the output side is down 16.7%.

Two other tiers go unmentioned. Work you aren't waiting on can go through batch processing at half price, and repeated long contexts that hit the cache are billed at $0.15 per million tokens, a tenth of the input rate.

$9.0 → $7.5
Per million output tokens, 3.5 Flash → 3.6 Flash (input holds at $1.50)
$0.75 / $3.75
Batch input / output pricing, half the standard rate
$0.15
Cache hit price, per million tokens
$1.00
Cache storage, per million tokens per hour
As we evaluated models for Figma Make, we wanted balance across quality, speed, and cost. Gemini 3.6 Flash hits that sweet spot, letting us explore and iterate on prototypes far faster while holding onto design quality. Matt Colyer, Director of Product at Figma (quoted on Google Blog)

Google also name-checks two customers, Hebbia and Harvey, saying the model is especially strong on multimodal knowledge work such as document parsing, chart and data analysis, and report drafting.

Scores

3.6 Flash beats last generation's flagship, 3.1 Pro, on all four benchmarks

In Google's four-panel bar chart, the pale bar on the far left is Gemini 3.1 Pro — the same preview model that remains the newest thing on the Pro line. The body of the launch page never discusses that column.

Bar chart of Gemini 3.6 Flash across four agentic benchmarks
Four agentic benchmarks; each cluster shows 3.1 Pro, 3.5 Flash, and 3.6 Flash. The dark blue 3.6 Flash bar on the right leads all four. Source: Google Blog
BenchmarkWhat it tests3.1 Pro3.5 Flash3.6 Flash
DeepSWE v1.1Long-horizon software engineering: read a real codebase, edit code, run tests12%37%49%
MLE-BenchMachine learning research (Google labels it ML Research)42.6%49.7%63.9%
GDPVal-AA v2Everyday tasks from real occupations, scored as Elo rather than percentage96513491421
OSWorld-VerifiedComputer use: clicking through a real system, opening apps, finishing tasks76.2%78.4%83.0%

That GDPVal-AA v2 row is Elo, not a percentage. Artificial Analysis describes the metric as "agentic real-world work tasks, (Elo−500)/2000."

Artificial Analysis rates 3.6 Flash (high thinking) at 50 on its Intelligence Index, a composite of nine evaluations.

The cheapest tier

3.5 Flash-Lite is the fastest model in the 3.5 family — and pricier than the one it replaces

The second release, Gemini 3.5 Flash-Lite, is positioned for low latency and high throughput, with agentic search and document processing called out as target workloads. Citing Artificial Analysis, Google says it produces 350 output tokens per second, the fastest in the 3.5 family.

Google's latency comparison: on the same high-frequency task, 3.5 Flash-Lite finishes visibly ahead of 3.5 Flash. Source: Google Blog
Two benchmarks comparing 3.5 Flash-Lite with 3.1 Flash-Lite
Against the previous 3.1 Flash-Lite. Left: Terminal-bench 2.1 rises from 31.0% to 54.0%, testing whether a model can finish jobs from the command line. Right: GDPVal-AA v2 climbs from 642 to 1140 (Elo). Source: Google Blog

Google adds one result that isn't in the chart: long-context retrieval on GDM-MRCR v2 improves from 60.1% to 72.2%, testing whether a model can pull the right passages out of extremely long material.

Two benchmarks comparing 3.5 Flash-Lite with 3 Flash
Against the older 3 Flash. Left: SWE-Bench Pro (Public), 49.6% vs 54.2%. Right: OSWorld-Verified, 65.1% vs 74.0%. Google's takeaway: work currently running on 2.5 Flash and 3 Flash can move over to this one. Source: Google Blog

What the launch page leaves out: it costs more than its predecessor

On price, the launch page says it is "priced at $0.30 / $2.50, with quality significantly better than 3.1 Flash-Lite and compelling value." Missing from that sentence: what 3.1 Flash-Lite cost. From the official pricing page:

3.1 Flash-Lite3.5 Flash-LiteChange
Input (text / image / video)$0.25$0.30+20%
Output$1.50$2.50+66.7%
Input (audio)$0.50$0.30−40%
Terminal-bench 2.131.0%54.0%+23 pts
GDPVal-AA v2 (Elo)6421140+498
Price check

Text output is up by two-thirds; audio input, going the other way, is down by 40%. The launch page sticks to "compelling value" throughout and never mentions the change in absolute price.

Its other selling point is adjustable thinking levels within a single model. Set it to minimal or low for high-frequency work when latency and cost matter; dial it up for sub-agent loads that need multi-step decomposition. At every thinking level, Google says, it beats 3.1 Flash-Lite by a wide margin. Computer use is now a built-in tool as well.

…the speed and cost efficiency Ashler needs for agentic retrieval and tool-calling. It excels at analyzing fragmented infrastructure records while holding context across hundreds of deliverables, supporting our users as they build the next generation of power grids, transit networks, and data centers. Connor Goggins, Co-founder and CTO of Ashler (quoted on Google Blog)

3.5 Flash-Lite is also rolling out into Google Search.

The specialist

3.5 Flash Cyber hunts vulnerabilities — and isn't for sale

The third release, Gemini 3.5 Flash Cyber, is fine-tuned from 3.5 Flash for one job: find security flaws in software, confirm they're actually exploitable, then write the patch. It costs less per token than larger security models, and it isn't sold publicly.

DeepMind's companion blog gives the backdrop: AI now finds vulnerabilities faster than existing systems can patch them, so defenders need something capable, cheap, and deployable at scale. Google says Flash Cyber's per-token price undercuts larger security models.

The idea: swap one expensive call for many cheap ones

Finding deep vulnerabilities means searching an enormous space of execution paths — the same function branches thousands of ways depending on inputs and call order. Expecting one giant model call to reason through all of them, DeepMind writes, is the bottleneck.

Google's answer is to call a small model many times. Inside CodeMender, its code security agent, multiple Flash Cyber agents each sweep a batch of code paths, and the results are synthesized into one report.

InputA large codebase
thousands of execution paths
Parallel scanMultiple Flash Cyber agents
each analyzing a batch of paths
SynthesisMerged into a single
high-quality report

DeepMind adds that as the number of calls goes up, Flash Cyber keeps turning up new code paths and new vulnerabilities. Because it's cheap, the whole setup can live inside high-frequency scanning, pre-release checks, and per-commit pipelines.

The full CyberGym ranking

Bar chart comparing five systems on CyberGym
CyberGym results: the dark blue bar is Gemini 3.5 Flash Cyber in CodeMender (max 5 model calls) at 83.2%, fourth of five. The y-axis starts at 60%, not 0. Source: Google Blog

CyberGym tests AI agents against hundreds of real software vulnerabilities. The launch page describes the result as reaching "frontier-level" performance; DeepMind's phrasing is "competitive performance against significantly larger models." Ranked by the numbers on the chart:

RankSystemCyberGym
1stGPT-5.5-Cyber in OpenAI Agent85.6%
2ndMythos 5 in Anthropic Agent83.8%
3rdGPT-5.6 Sol in OpenAI Agent83.6%
4thGemini 3.5 Flash Cyber in CodeMender (max 5 model calls)83.2%
5thMythos Preview in Anthropic Agent83.1%

Fourth place sits 0.1 points above fifth and 2.4 points below first. Google's own bar carries a qualifier: a maximum of five model calls. DeepMind explains that the score comes from configuring CodeMender to make at most five calls per final report, and a footnote states that competitor scores are taken from each vendor's self-reported numbers.

Related on this site
GPT‑5.6 launched, sort of
GPT-5.6 Sol, third on this leaderboard, got its own breakdown here on June 27, 2026.

The V8 run: 55 vs 47 vs 36

V8 is Chrome's JavaScript engine, which DeepMind calls extremely complex. Given a fixed number of calls, here's how many unique, confirmed issues each model surfaced:

55
Unique confirmed issues found by 3.5 Flash Cyber
47
Found by the mainline 3.5 Flash
36
Found by Claude Opus 4.6
10
Found only by Flash Cyber, missed by both others

Google explains why "unique" matters: weaker security models tend to get stuck in a loop, reporting the same issue over and over while missing the ones that count. Stronger models cast a wider net, which is what produces more distinct findings.

The other two runs are scored pass@1. On an independent evaluation built by Google's Big Sleep team — which deliberately targets the hardest critical vulnerabilities in complex codebases like Chrome and Safari — Flash Cyber clearly outperforms both mainline 3.5 Flash and 3.6 Flash. On Chrome's production commit-scanning pipeline it improves substantially over 3.5 Flash; none of the vulnerabilities in that set have ever been publicly disclosed, which Google says keeps the benchmark free of data contamination for Gemini and competitors alike. Early feedback from security engineering testers at Wiz and Cloud CISO also confirmed the gain over mainline 3.5 Flash.

DeepMind spells out the conditions for both runs: Google's own testing was done as a stress test with safety guardrails turned off. And in the Chrome commit-scanning run, competitor models newer than Claude Opus 4.6 refused the task because of their built-in safety guardrails, so they were left off that chart.

In the field: a working exploit chain in two hours

Real case

Google Cloud's vulnerability research team pointed Flash Cyber at their own systems. Within two hours, it found remote code execution flaws across several public APIs, plus a memory corruption bug in a sensitive production service.

It then produced a 100% reliable remote code execution exploit that bypassed standard defenses like ASLR and W^X — the first randomizes where a program sits in memory on each run, the second forbids any memory region from being both writable and executable.

Running inside CodeMender, Flash Cyber is already finding and fixing bugs across Google's internal Chrome, Android, Cloud, Ads, and YouTube codebases. Google claims an advantage in training data that others don't have: OSV.dev, an open source vulnerability database with more than 700,000 records, plus over a decade of accumulated OSS-Fuzz fuzzing results. That, Google says, means the model isn't stuck on synthetic exercises — it learns to drive industry-standard tools, read through million-line projects like Chromium, and work through security tasks that demand hours of sustained analysis.

Given the dual-use nature of the technology, DeepMind writes, Google made a deliberate choice about how to ship it: no public sale. It will go to governments and trusted partners through CodeMender as a limited pilot, expanding gradually from there. The launch page uses the future tense; nothing is open yet. Separately, Google is opening CodeMender's underlying capabilities to ordinary customers through the Gemini Enterprise Agent Platform, using generally available Gemini models.

Getting access

Where you can use these two today

3.6 Flash and 3.5 Flash-Lite are both available from day one:

  • Developers: the Gemini API, reachable through Google AI Studio and Android Studio; 3.6 Flash is also in Google Antigravity.
  • Enterprises: the Gemini Enterprise Agent Platform, with 3.6 Flash selectable in the Gemini Enterprise app as well.
  • Everyone else: pick it in the Gemini app; 3.5 Flash-Lite is rolling out into Google Search.

Computer use — letting the model drive a machine's interface directly — is now a built-in client-side tool through the Gemini API and Gemini Enterprise, supported by both new models.

On safety, 3.6 Flash tightens defenses in two areas: chemical, biological, radiological, and nuclear risks (CBRN), and cyberattack misuse. Google says this makes the model considerably harder to jailbreak while also training it to refuse legitimate requests less often. Both models have public model cards with the details.

🧰 Quick start · Gemini 3.6 Flash
Price$1.50 input / $7.50 output per million tokens; batch halves that to $0.75 / $3.75; cache hits $0.15; Google AI Studio includes a free tier
BarrierSelect the model on the Google AI Studio website and start chatting, or write code against the Gemini API; 3.5 Flash-Lite is in the same picker
Source
Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash CyberGoogle·blog.google·2026-07-21
Editor's note
All charts and videos come from Google's launch page, with the eight Demo Lab clips found in its media carousel. Two things here do not come from the launch page body: (1) historical pricing for 3.5 Flash and 3.1 Flash-Lite, batch and cache rates, and Pro line version status, all taken from the official Gemini API pricing page; (2) the Artificial Analysis Intelligence Index score of 50 and its nine components, taken from that site's evaluation page. The CyberGym ranking follows the numbers printed on Google's chart.