Google's summer of 2026 wasn't about building the smartest AI on paper — it was about building the AI that gets work done fastest, cheapest, and in the background while you sleep. Between July 16 and July 21, 2026, Google shipped three new Gemini models, a 24/7 AI agent that runs without your laptop open, a research organization feature, and a text-to-video tool — all aimed at the same proposition: stop paying for raw intelligence you don't use, and start paying for the right tool on the right job.
The headline release is Gemini 3.6 Flash (with 3.5 Flash-Lite and the restricted 3.5 Flash-Cyber), launched July 21, 2026. It uses ~17% fewer output tokens than the model it replaces, costs $7.50 per million output tokens (down from $9.00), and posted higher scores on coding, long-context, and computer-use benchmarks. The same day, Google shipped Flash-Lite at $0.30/$2.50 per million tokens for high-volume repetitive work, and Flash-Cyber for government vulnerability finding. Gemini Spark — your always-on personal agent — expanded to the $20/month Google AI Pro tier that same week. And Gemini Notebook Collections rolled out to tame your research pile.
Here's what each one does, what it costs, and whether it matters for how you work.
Last verified: July 30, 2026 · Pricing and availability change quickly — verify at Google's official pages before purchasing.
- Gemini 3.6 Flash: new default workhorse, $1.50/$7.50 per M tokens, 17% fewer output tokens
- Gemini 3.5 Flash-Lite: cheapest tier, $0.30/$2.50 per M tokens, 350 tokens/sec
- Gemini 3.5 Flash-Cyber: security-only, restricted to governments/CodeMender — no public access
- Gemini Spark: 24/7 personal agent, now on AI Pro ($19.99/mo) and AI Ultra ($99.99+/mo)
- Gemini Notebook Collections: organize notebooks into flexible groups (rolling out)
- Google Vids + Gemini Omni: generate and edit videos by typing what you want changed
What Did Google Release on July 21, 2026? Three New Gemini Models
Google released three models on a single day: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber, according to Google's official announcement (Confirmed).
| Model | Input / Output (per 1M tokens) | Best for | Available to |
|---|---|---|---|
| Gemini 3.6 Flash | $1.50 / $7.50 | General coding, writing, multimodal, computer use | All developers + consumers |
| Gemini 3.5 Flash-Lite | $0.30 / $2.50 | High-volume repetitive tasks, data extraction | All developers + consumers |
| Gemini 3.5 Flash-Cyber | Not published (pilot) | Finding + patching security vulnerabilities | Governments + trusted partners only |
Notably, Gemini 3.5 Pro was not released. Google confirmed it's "testing with partners" and that pre-training has already begun on Gemini 4. The signal is deliberate: the workhorse Flash tier keeps iterating while the flagship Pro tier waits.
What makes Gemini 3.6 Flash different?
Gemini 3.6 Flash is the new default in the Gemini family — the model Google wants everyone using by default for everyday work. The model card from Google DeepMind confirms a 1 million-token context window and a March 2026 knowledge cutoff.
The win isn't raw intelligence — it's token efficiency. Google's own figures say 3.6 Flash uses roughly 17% fewer output tokens than its predecessor (3.5 Flash) while scoring higher across DeepSWE (49% vs 37%), MLE-Bench (63.9% vs 49.7%), and OSWorld-Verified computer use (83.0% vs 78.4%), per AI Release Tracker and AIToolsReview. For your budget, that means fewer tokens spent to reach the same answer.
When to use Gemini 3.5 Flash-Lite instead?
Think of Flash-Lite as the model for the work you do a thousand times a day — sorting incoming messages, reading documents, answering simple questions, extracting product features from listings. Google measured it at 350 output tokens per second, per The Keyword announcement. It roughly doubled 3.1 Flash-Lite's Terminal-Bench 2.1 score (54% vs 31%), per AIToolsReview.
You don't need a surgeon to refill your water glass. Flash-Lite is fast, reliable, available all day — the model you point at volume work, not complex reasoning.
What is Gemini 3.5 Flash-Cyber?
Flash-Cyber is a security-specialised model fine-tuned on top of 3.5 Flash to find, validate, and patch software vulnerabilities. It only runs inside Google's CodeMender agent, where multiple Cyber-model sub-agents collaborate to produce a single vulnerability report. It's not available to the public, and is gated to governments and trusted partners under a limited-access pilot program, per Google DeepMind's model page.
It still matters for context: Google is building specialised AI for serious problems — not just chatty assistants. If your business is consumer-facing, you'll never touch Flash-Cyber directly, but the defensive security it enables could protect the platforms you build on.
How Much Do the New Gemini Models Cost?
Pricing is the real story of this release. For developers building production apps on the Gemini API:
| Model | Input (per 1M tokens) | Cached input | Output (per 1M tokens) | Context |
|---|---|---|---|---|
| Gemini 3.6 Flash | $1.50 | $0.15 | $7.50 | 1M |
| Gemini 3.5 Flash (superseded) | $1.50 | $0.15 | $9.00 | 1M |
| Gemini 3.5 Flash-Lite | $0.30 | — | $2.50 | 1M |
Figures from Google's official Gemini API pricing page and verified against FelloAI's pricing tracker as of July 2026. The output price for 3.6 Flash dropped from $9.00 (3.5 Flash) to $7.50 — a 17% cost reduction on the same job.
For consumer subscriptions:
- Google AI Plus: $4.99/mo (cut from $7.99 on June 8, 2026)
- Google AI Pro: $19.99/mo (now includes Gemini Spark access, expanded July 2026)
- Google AI Ultra: from $99.99/mo; the top Ultra tier is $199.99/mo with up to 20x Pro limits
Consumer pricing verified from Suprmind's pricing aggregator and FelloAI, July 2026.
What Is Gemini Spark? The 24/7 Agent That Keeps Working After You Close Your Laptop
Gemini Spark is Google's personal AI agent — not a chat window, but an agent that keeps working after you walk away. CEO Sundar Pichai described it as "your personal AI agent in Gemini app that helps you navigate your digital life, taking action on your behalf and under your direction," per Google's I/O 2026 keynote transcript (Confirmed).
The architecture matters:
- Runs on dedicated virtual machines on Google Cloud, so it stays on whether your laptop is open or not
- Powered by Gemini 3.5 and the Google Antigravity harness, which handles long-horizon tasks in the background
- Integrates with Google's own tools, with third-party integrations coming through MCP (Model Context Protocol)
- Available via the Gemini app, and soon through email and chat
The shift is the language itself. You don't tell Spark "do this one thing" — you tell it "watch this," "keep checking," "handle this every week without me asking again." That's the agentic shift everyone in AI keeps talking about: from commands to delegations.
What does Gemini Spark cost and who can access it?
- Started May 19, 2026 for trusted testers at Google I/O 2026
- U.S. beta opened to Google AI Ultra subscribers the week of May 25
- As of July 2026, expanded to Google AI Pro subscribers ($19.99/mo) — confirmed by MegaMobile Content
- Region availability: Supported everywhere Gemini Apps are available except the EEA, Nigeria, Switzerland, and the United Kingdom — verified against Google's Gemini Help Center (AI Agents Library)
Note: One source (Digital Applied) reported Spark launched on AI Pro and Ultra together at I/O; Google's Help Center and most reporting place Pro access as a July 2026 expansion. Treat the July expansion as Confirmed and the May overlap as Reported.
What Are Gemini Notebook Collections? The Organization Fix You Asked For
If you've used Gemini Notebook (formerly NotebookLM) heavily, you know the pain: dozens of notebooks with no real order, one flat chronological list. Collections fixes that.
Started rolling out July 21, 2026, per Android Authority, Collections let you group related notebooks the way you'd make a music playlist. A single notebook can belong to multiple Collections. No rigid folders — just flexible, semantically sensible groups.
The practical value: build a Collection for "content ideas," drop in every testimonial, case study, and call recap, then ask Gemini to synthesise them into one clean summary you can turn into next week's post. The feature is rolling out gradually and is currently limited to personal organisation — no support for sharing Collections yet.
For more on the broader NotebookLM platform, see our full guides on Gemini Notebook (formerly NotebookLM) in 2026 and how to organize Gemini Notebook Collections.
What Is Google Vids + Gemini Omni? Video Generation by Typing
Google Vids added Gemini Omni on July 16, 2026, giving you the ability to generate and edit high-quality video clips from a simple text prompt, optionally seeded with an image or sketch as a reference. Per Google's Keyword announcement and Google Workspace Updates:
- Generate a video by typing what you want to see, optionally adding a photo or rough sketch as a reference
- Edit by typing: "fix the color-grading," "restyle the visuals in anime," or "remove that New York siren in the background"
- The model supports step-by-step edits — you tweak without starting from scratch
- Every AI-generated clip carries an invisible SynthID watermark for content transparency
Availability: Google AI Pro and AI Ultra consumers, plus Workspace Business and Enterprise customers. Rollout is gradual (up to 15 days for visibility). Editing non-AI videos with Omni is not available at launch in the EEA, Switzerland, the UK, Texas, or Illinois.
If you want to embed AI video in your small business content without licensing separate AI tools, this brings video generation into the Workspace apps you already pay for. For a wider look at free AI video options, see our guide on how to use premium AI video models for free in 2026.
The Big Pattern: Efficiency Over Raw Intelligence
Step back and the July 2026 releases tell one story: AI's new bragging rights aren't about being smarter, they're about how little it takes to get the job done.
A year ago, every new model release was a brag about raw intelligence. Now the pitch is token efficiency (17% fewer output tokens), price-per-token cuts ($9 → $7.50 output), throughput (350 tokens/sec on Flash-Lite), and always-on persistence (Spark runs without you). That's a fundamentally different kind of race.
For a business owner, this maps to three concrete choices:
- High-volume repetitive work → Flash-Lite at $0.30/$2.50 per M tokens. Don't pay frontier prices for sorting and filtering.
- Coding, writing, multimodal → 3.6 Flash at $1.50/$7.50. Same context window, better benchmarks, 17% cheaper output.
- Things you keep forgetting to follow up on → Gemini Spark, the always-on agent. Pay $19.99/mo for delegation, not for a smarter answer to a one-shot question.
What This Means for You
You don't need the most powerful model available — you need the right tool for the right job. The three model tiers and Spark give you the dials:
- Fast and cheap (Flash-Lite) for the tasks you do a thousand times a day.
- Sharper reasoning (3.6 Flash) for the work that actually needs thinking.
- A background agent (Spark) for the follow-ups and monitoring you'd otherwise drop.
Pick one update from this release. Not all of them — one. Test it on one real workflow this week. The honest part: tools move fast and feel overwhelming. New models and agents ship almost weekly. You don't need to master all of it overnight. You just need to apply one thing consistently. For deeper reads on where Google's agentic platform is heading, see our guides on Google Antigravity for business automation, Gemini 3.6 Flash for AI agents, Gemini in Google Workspace, and setting up AI agents for productivity in 2026.
FAQ
Q: What did Google release on July 21, 2026? A: Three new Gemini models on a single day: Gemini 3.6 Flash (the new default workhorse, $1.50/$7.50 per million tokens), Gemini 3.5 Flash-Lite (the cheapest tier at $0.30/$2.50 per million tokens), and Gemini 3.5 Flash-Cyber (a security-specialised model restricted to governments and trusted partners through Google's CodeMender agent). Gemini 3.5 Pro was not released and remains in partner testing.
Q: How much does Gemini 3.6 Flash cost? A: Gemini 3.6 Flash costs $1.50 per million input tokens and $7.50 per million output tokens, per Google's official API pricing. That is a 17% reduction from the output price of the model it replaces (Gemini 3.5 Flash at $9.00 per million output tokens). It uses roughly 17% fewer output tokens on the same queries while scoring higher on coding, long-context, and computer-use benchmarks.
Q: What is Gemini Spark and how do I access it? A: Gemini Spark is Google's 24/7 personal AI agent that runs on dedicated Google Cloud virtual machines and continues working after you close your laptop. It was announced May 19, 2026 at Google I/O and expanded to Google AI Pro subscribers ($19.99/mo) in July 2026. Access requires being 18+, a personal Google Account, Keep Activity enabled, and using Spark in English through the Gemini web, mobile, or macOS app. It is not available in the EEA, Nigeria, Switzerland, or the United Kingdom.
Q: What are Gemini Notebook Collections? A: Collections is a new organizational feature rolling out to Gemini Notebook (formerly NotebookLM) starting July 21, 2026 that lets you group related notebooks the way you would build a music playlist. A single notebook can belong to multiple Collections. It's currently limited to personal organisation and does not support sharing yet — the rollout is gradual over up to 15 days.
Q: Can I use Google Vids and Gemini Omni for free? A: Google Vids with Gemini Omni is available to Google AI Pro ($19.99/mo) and AI Ultra subscribers, and to Google Workspace Business and Enterprise customers. It is not available on a free tier. Editing non-AI videos with Omni is not available at launch in the EEA, Switzerland, the UK, Texas, or Illinois. Every AI-generated clip carries an invisible SynthID digital watermark for transparency.
Q: Was Gemini 3.5 Pro released in July 2026? A: No. Google confirmed in its July 21, 2026 announcement that Gemini 3.5 Pro is "currently testing with partners" with no general availability date. The same announcement stated that pre-training has already begun on Gemini 4. The signal is that Google is iterating on its workhorse Flash tier while the flagship Pro tier waits until it can compete with the current frontier.
Discussion
0 comments