Google's July 2026 Gemini updates landed four major changes at once: a smarter workhorse model (Gemini 3.6 Flash), a budget-tier speed model (Gemini 3.5 Flash-Lite), a full rebrand of NotebookLM into Gemini Notebook with code execution, and a conversational video editing model (Gemini Omni Flash). For anyone using AI for daily work, the practical takeaway is simple — you get better intelligence at lower cost, a research tool that now runs its own analysis, and the ability to edit video by typing instructions in plain English.
Last verified: 2026-07-29
- Gemini 3.6 Flash and 3.5 Flash-Lite are both GA as of July 21, 2026
- NotebookLM was renamed to Gemini Notebook on July 16, 2026
- Gemini Omni Flash launched May 19, 2026 and is now in Google Vids
- Pricing and model availability change often — verify before committing
What are the Google Gemini updates from July 2026?
The July 2026 Gemini update delivered four changes: Gemini 3.6 Flash (a new workhorse model with a March 2026 knowledge cutoff and 17% fewer output tokens), Gemini 3.5 Flash-Lite (the fastest, cheapest model for high-volume tasks), NotebookLM becoming Gemini Notebook (with collections, code execution, and cross-device sync), and Gemini Omni Flash gaining conversational video editing inside Google Vids. Google announced the models on July 21, 2026, and the NotebookLM rename on July 16, 2026.
This is not a minor refresh. Three new model variants shipped on the same day, the research product got a full rebrand plus a compute layer, and video editing became a chat-driven workflow. If you use Gemini for any kind of recurring work — content, coding, research, customer support — at least one of these changes directly affects your cost or capability.
What is Gemini 3.6 Flash and why does it matter?
Gemini 3.6 Flash is Google's new general-purpose workhorse model, released as generally available (GA) on July 21, 2026. It is built on Gemini 3.5 Flash and delivers better coding, knowledge work, and multimodal performance while using fewer tokens — up to 17% fewer output tokens for equivalent tasks than its predecessor. (Google DeepMind Model Card)
The four numbers that matter:
| Metric | Gemini 3.5 Flash (previous) | Gemini 3.6 Flash (new) |
|---|---|---|
| Output price (per 1M tokens) | $9.00 | $7.50 |
| Input price (per 1M tokens) | — | $1.50 |
| Output tokens for equivalent tasks | baseline | up to 17% fewer |
| Knowledge cutoff | January 2025 | March 2026 |
| OSWorld-Verified (computer-use) | 78.4% | 83% |
| Context window | 1M tokens | 1M tokens |
Source: Google's official announcement and DeepMind model card
The token-efficiency improvement is the real story. A 17% cut in output tokens, layered on top of a lower per-token price, compounds. If your current Flash-tier spend is meaningful, the effective cost reduction for agentic workloads is larger than the headline price drop suggests. The knowledge cutoff jump from January 2025 to March 2026 also closes a gap that mattered for anything touching recent library versions, current APIs, or world events from the last 15 months.
Note: Google cautions that while the knowledge cutoff is March 2026, users may still experience the model's knowledge being limited to January 2025 in some domains, consistent with the Gemini 3 Model Family. (DeepMind Model Card)
Google also deprecated the temperature, top_p, and top_k sampling parameters with this release, which matters if you have existing prompts that rely on them. (Gemini API Changelog)
What is Gemini 3.5 Flash-Lite and when should you use it?
Gemini 3.5 Flash-Lite (also released GA on July 21, 2026) is Google's fastest, most cost-effective model in the 3.5 series, built for high-volume, latency-sensitive tasks. It costs $0.30 per million input tokens and $2.50 per million output tokens — roughly 94% cheaper than Gemini 3.6 Flash on the same tasks at a per-call level. (OpenRouter model page, Google's announcement)
The practical routing rule: start with Flash-Lite for simple, high-throughput work, and step up to 3.6 Flash only when the output actually breaks. Flash-Lite handles straightforward extraction, translation, classification, and data processing at scale. In independent testing, it answered simple coding, reasoning, and JSON extraction tasks correctly while streaming at over 400 tokens per second. (AIReiter)
Where Flash-Lite falls short: deep reasoning, multi-step agentic workflows, and complex coding. Its intelligence index sits at 37 (vs. 50 for 3.6 Flash), and time-to-first-token runs around 6 seconds when thinking is enabled — on the high side for live chat but fine for background jobs. (Artificial Analysis via OpenRouter)
A worked cost example: if you process 50,000 support tickets per month at roughly 1,000 input tokens and 200 output tokens per ticket, Flash-Lite costs about $40/month in raw model cost. The same volume on 3.6 Flash would cost closer to $150. (eesel AI)
For a deeper comparison of when to pick each model, see our guide on Gemini 3.6 Flash vs 3.5 Flash-Lite: which Google model should you actually use.
How does the Planner-Executor pattern work with the new Gemini models?
One of the highest-leverage workflows the new model lineup enables is a two-model orchestration: run Gemini 3.6 Flash as the planner (intelligent, good at breaking down complex tasks) and a fleet of Gemini 3.5 Flash-Lite instances as executors (fast, cheap, good at individual steps). This cuts total cost dramatically while keeping quality high on the planning layer.
The pattern works like this:
- Plan with 3.6 Flash: Send the complex task prompt to 3.6 Flash. It decomposes the work into sub-tasks with clear, self-contained instructions for each.
- Execute with Flash-Lite: Send each sub-task to a Flash-Lite instance. Each one runs at 350+ tokens per second at $0.30/$2.50 per million tokens.
- Verify with 3.6 Flash: Collect the Flash-Lite outputs and send them back to 3.6 Flash for review. It checks for errors, inconsistencies, or missing pieces.
- Fix or finalize: If errors are found, re-execute only the failed sub-tasks with Flash-Lite. If everything passes, deliver the final output.
This approach can save 70–80% on total token cost compared to running everything through 3.6 Flash alone, with minimal quality loss on the execution layer. For a detailed implementation guide, see our Planner-Executor two-model orchestration playbook.
What happened to NotebookLM? Meet Gemini Notebook
Google renamed NotebookLM to Gemini Notebook on July 16, 2026. It remains the same standalone product at the same web address (notebooklm.google), and existing notebooks and shared links continue to work through automatic redirects. The rename brings NotebookLM under the unified Gemini brand, alongside the earlier rebranding of Bard to Gemini and Duet AI to Gemini for Workspace. (Google's announcement, Workspace Updates)
But the rename is the least interesting part. The capability upgrades matter more:
Collections. You can now group related notebooks together. If you have 10 notebooks for one project, you can organize them in one place with emoji labels for faster navigation.
Cloud computer (code execution). This is the big one. Each notebook now has a secure, sandboxed cloud computer that lets Gemini Notebook write and run code on your behalf — cleaning inconsistent data, performing calculations, generating charts, and doing real analysis grounded in your uploaded sources. Google upgraded the underlying models to Gemini 3.5 and Antigravity in June 2026, and the code execution capability rolled out with the rename. (Google's research upgrade post, June 8, 2026)
Availability of the cloud computer feature: live for Google AI Ultra users and Workspace users at launch, rolling out to Pro users on the web over the following weeks.
Cross-device sync. Notebooks now follow you across surfaces. You can create a notebook in the Gemini app, open it in Gemini Notebook on the web, and they stay in sync. Google has also said notebooks will come to AI Mode in Search soon.
The practical workflow that makes Gemini Notebook different from a chatbot: instead of asking generic questions, you load it with your own documents — training materials, call transcripts, SOPs, research papers — and ask it to analyze them grounded in those sources. It will not hallucinate facts from outside your uploaded materials. The code execution means it can now run real calculations on the data in those documents and produce verified results, not just text summaries.
If you want to go deeper on building a research workflow with Gemini Notebook, our guide to turning NotebookLM into a personal research engine walks through the full architecture.
What is Gemini Omni and how does conversational video editing work?
Gemini Omni is Google's multimodal video generation and editing model, first announced at Google I/O 2026 on May 19, 2026. The first model in the family, Gemini Omni Flash, can combine text, images, video, and some audio references into a single video workflow, and then keep editing the same project through natural-language conversation — not just one-shot generation. (Google's Gemini Omni announcement)
In July 2026, Google brought Gemini Omni into Vids (its video creation tool in Workspace), letting users edit existing videos conversationally. You can now tell Vids to "fix the color-grading," "restyle the visuals in anime," or "remove that background siren" with a simple text instruction. (Google Workspace Updates, July 2026)
The Omni Flash model generates 10-second clips with synchronized audio and supports conversational editing across multiple turns. It launched in the Gemini app, Google Flow, and YouTube Shorts/Create for Google AI Plus, Pro, and Ultra subscribers worldwide. The developer API and enterprise rollout followed in subsequent weeks. (Nerova analysis)
All videos created with Omni include SynthID watermarking — an imperceptible digital signature embedded in the video frames — so AI-generated content can be identified. (Google's announcement)
For a comparison of where Gemini Omni fits among other AI video tools, see our tested and ranked guide to the best free AI video generators in 2026.
What do the Gemini updates mean for small businesses and builders?
The July 2026 updates change three things for anyone using AI in their daily work:
1. Your AI budget goes further. Between 3.6 Flash's 17% token efficiency gain and Flash-Lite's $0.30/$2.50 pricing, the cost per task is dropping meaningfully. If you are running hundreds or thousands of AI calls per day for content generation, customer support triage, or data extraction, switching to the right model tier can cut your monthly spend by 50–80% without changing your workflows. Start with Flash-Lite, step up to 3.6 Flash only when quality drops. This is the same routing approach we recommend in our Gemini 3.6 Flash switch-or-wait cost routing guide.
2. Your research tool now actually computes. Gemini Notebook's cloud computer means you are no longer limited to asking "summarize these documents" — you can ask "analyze these sales call transcripts, calculate the win rate by deal size, and tell me which pricing objections appear most often in lost deals." It runs the code. It shows its work. And it stays grounded in your uploaded sources.
3. Video creation is no longer a specialized skill. Gemini Omni's conversational editing in Vids means a small business owner can take product footage, ask for a 15-second ad with a specific benefit highlighted, and refine it by saying "make the text bigger" or "swap the intro shot" — no video editing software, no timeline, no learning curve. Google also brought Gemini Omni into Google Vids for Workspace users in July 2026, making it available to business teams. For the full scope of what changed across Google Workspace this month, see our Google Gemini Workspace updates July 2026 guide.
What this means for you
- If you build agents or automation: Test the 3.6 Flash + Flash-Lite Planner-Executor combo. The cost savings on execution are real and measurable.
- If you do research or content work: Move your source documents into Gemini Notebook and try the code execution feature. The ability to get grounded analytical output — not just summaries — is a genuine step change.
- If you create marketing content: Try Gemini Omni in Vids for quick product or social video. The conversational editing removes the need for traditional editing software for simple assets.
- If you are waiting for Gemini 3.5 Pro: It is still delayed with no firm date. Google publicly teased Gemini 4 alongside the July 21 release, suggesting Pro-tier gains may land with the next major version. Meanwhile, 3.6 Flash is your best workhorse. See our Gemini 3.5 Pro delay guide for what to do meanwhile.
FAQ
Q: When were Gemini 3.6 Flash and 3.5 Flash-Lite released? A: Both models shipped as generally available (GA) on July 21, 2026, alongside a third model called Gemini 3.5 Flash Cyber (a cybersecurity-tuned variant). They are available in Google AI Studio and Vertex AI. (Google announcement)
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. The previous Gemini 3.5 Flash charged $9.00 per million output tokens. The effective cost drop is larger than the price cut because 3.6 Flash uses up to 17% fewer output tokens for equivalent tasks. (DeepMind Model Card)
Q: What is the difference between Gemini 3.6 Flash and Gemini 3.5 Flash-Lite? A: 3.6 Flash is the intelligent workhorse — better at coding, agentic tasks, and multimodal work, with a higher intelligence index (50 vs. 37). Flash-Lite is the speed-and-cost tier — 94% cheaper per call for simple tasks, streaming at 350+ tokens per second, but limited on deep reasoning. Use Flash-Lite for high-volume simple work, 3.6 Flash for complex multi-step tasks. (eesel AI)
Q: Is NotebookLM going away? A: No. NotebookLM was renamed to Gemini Notebook on July 16, 2026. It is the same product at the same web address. Existing notebooks and shared links continue to work through automatic redirects. No admin action is required for Workspace customers. (Google Workspace Updates)
Q: Can Gemini Notebook run code? A: Yes. Each notebook now has a secure, sandboxed cloud computer that can write and run code for data analysis, calculations, and chart generation, grounded in your uploaded sources. This feature is live for Google AI Ultra and Workspace users, and rolling out to Pro users on the web. (Google's research upgrade post)
Q: What is Gemini Omni and who can use it? A: Gemini Omni is Google's conversational video generation and editing model. The first model, Gemini Omni Flash, launched on May 19, 2026 for Google AI Plus, Pro, and Ultra subscribers. In July 2026, Google brought Omni into Vids for Workspace users, letting teams edit existing videos conversationally. An API and enterprise rollout followed. (Google's Omni announcement)
Q: What is the knowledge cutoff for the new Gemini models? A: Gemini 3.6 Flash and 3.5 Flash-Lite both have a knowledge cutoff of March 2026, up from January 2025 on the previous Gemini 3.5 Flash. However, Google notes that in some domains the model's knowledge may still be limited to January 2025. (DeepMind Model Card)
Discussion
0 comments