This Week in AI: Google's Gemini 4 Carbon, Genie 4, and a Cloud Earnings Reality Check

Google owned this week, twice over. On October 27, DeepMind unveiled Genie 4, a world model that generates sound along with the scene. On October 28, Google announced Gemini 4 Carbon, its strongest model yet, and handed it first to cyber defenders rather than to everyone. Around those launches, the money story caught up with the model story: Microsoft and Alphabet both reported cloud growth that would be extraordinary in any other industry, and both are spending at a pace that made even good results feel expensive. Here is what happened, with every claim verified against primary or reputable sources.

Key Takeaways

Google announced Gemini 4 Carbon on October 28, scoring 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, and ranking first on the Artificial Analysis Intelligence Index v5 at 59. (Google)

Google DeepMind unveiled Genie 4 on October 27, its first real-time world model to generate audio natively, with scenes that stay consistent for about two minutes, double the limit of Genie 3. (Unite.AI)

Alphabet reported third-quarter revenue of $127.6 billion, up 25%, with Google Cloud up 103% to $30.7 billion, while Microsoft reported fiscal first-quarter revenue of $92.3 billion, up 19%, with Azure revenue of $33.2 billion, up 48%. (Fox Business)

OpenAI and ServiceNow expanded their partnership on October 28, launching a ServiceNow MCP plugin in the OpenAI marketplace so teams can act on workflows from inside ChatGPT. (Unite.AI)

Liquid AI released LFM3 on October 27, six open-weight models of about 2 billion parameters each, branching from one 1.7-billion-parameter base, with the text model fitting in about 1 GB at 4-bit precision. (Unite.AI)

1. Gemini 4 Carbon: Google's Strongest Model Goes to Cyber Defenders First

Google announced Gemini 4 Carbon on October 28, calling it its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities, reports Google's official blog. Carbon arrives just two weeks after Gemini 4 Argon, and Google is deliberately not giving it to everyone at once. Starting this week, a subset of cyber defenders gets access first, with wider access phased through Google's Fairwind Program before developers, businesses, and consumers get it. Google says it is engaged in the US government's voluntary pre-release model testing process while it iterates on guardrails.

The cybersecurity numbers are the headline, and Google is specific about them. Evaluated without guardrails on external benchmarks, Carbon scored 88.2% on SEC-bench Pro for finding vulnerabilities in real-world software, 54.8% on ExploitGym for validating them through exploitation, and 71.0% on CWE-bench for patching them at the root cause. It also ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59. On internal benchmarks, Google says Carbon more than doubles Argon's success rate at generating secure code, and its Chrome team reported Carbon performing more than 10 times better than another lab's large frontier model at generating high-quality patches. These are company-reported figures, not independent tests, and the external benchmark scores were measured without guardrails, which is exactly how a defender would want a security tool measured, and exactly why the rollout is gated.

Carbon is also built for long, stubborn engineering work. Google says its internal teams have seen it run for days, persisting through dead ends to root-cause and patch subtle bugs across large codebases. On WebDev Arena it gained 89 Elo points to reach fourth place at 1768. Pricing matches Argon's introductory rate: $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off the input price, rising to $4 and $20 when the introductory period ends. The message is clear: the frontier model of the week is not a chatbot demo. It is a security tool, priced like infrastructure, and released like one too, to the people who defend systems first.

2. Genie 4: A World Model That Finally Has Sound

A day before Carbon, Google DeepMind unveiled Genie 4 on October 27, a world model for real-time interactive simulation that generates audio natively alongside the visuals, reports Unite.AI, summarizing DeepMind's announcement. Type a prompt and you get a scene you can walk through with your keyboard, then steer while it runs: change the weather, move the camera, introduce a character, and the world updates without restarting. The sound is the new part. Music, ambient noise, and character dialogue are generated in the same pass as the visuals, not layered on afterward, which DeepMind says makes Genie 4 its first real-time world model to generate audio natively.

Under the hood, two Gemini models do the steering. Gemini Omni anchors how people and places look and move, while Gemini 3.8 Flash acts as a guiding agent inside the loop, breaking a goal into steps and checking frames to confirm each one landed. The practical limit is about two minutes of consistent simulation, double the one-minute limit of Genie 3, after which object permanence and fine details can drift. DeepMind is frank that this is an early, experimental model, not a finished product.

Access is narrow for now. Genie 4 is a research preview for trusted testers across creative industries, journalism, education, gaming, and engineering, with Google AI Ultra subscribers in Project Genie and early-access partners in the Gemini API and AI Studio next in line. Every video and audio clip carries Google's SynthID watermark. The strategic read matters more than the demos: DeepMind frames world simulation as a safe, infinitely variable training ground for robots and agents. If agents are going to act in the physical world, they need somewhere cheap and safe to practice first, and Genie 4 is Google's bid to own that practice ground.

3. Microsoft and Alphabet Earnings: Cloud Demand Doubles, and So Does the Bill

Both cloud giants reported on October 28, and the numbers tell one story with two endings. Alphabet reported third-quarter revenue of $127.6 billion, up 25% from a year earlier, while Microsoft posted fiscal first-quarter revenue of $92.3 billion, up 19%, reports Fox Business. The cloud lines are where AI demand shows up. Google Cloud revenue more than doubled, up 103% to $30.7 billion, and Microsoft disclosed Azure revenue in dollars for the first time: $33.2 billion, up 48%. Microsoft Cloud revenue, a broader category, rose 30% to $63.9 billion.

Alphabet CEO Sundar Pichai said on the earnings call that AI infrastructure and solutions are the major drivers of Cloud growth and that demand is substantial, adding that the company is working to bring more capacity online. Alphabet projected capital expenditures of between $195 billion and $205 billion this year, with spending expected to rise significantly in 2027. That spending is the second ending. Alphabet's shares fell about 10% in after-hours trading despite the cloud beat, as investors weighed negative free cash flow against the buildout, while Microsoft shares rose about 2% after beating expectations. The week also carried a reminder from last week's roundup: Anthropic's IPO filing disclosed compute costs of $8.9 billion in the first half of 2026 alone. Demand is not the question anymore. Whether the buildout pays for itself, and how fast, is.

4. OpenAI and ServiceNow: Workflows Move Into ChatGPT

OpenAI and ServiceNow announced an expanded partnership on October 28 that brings more OpenAI models and agents into ServiceNow products and connects ServiceNow workflows with ChatGPT and Codex, reports Unite.AI. The centerpiece available now is a ServiceNow MCP plugin in the OpenAI marketplace. An IT team member can find similar incidents, assess the risk of a change, and update a ServiceNow record without leaving the ChatGPT conversation. Customers connect ChatGPT to their own ServiceNow instance and choose which capabilities to expose, and ServiceNow's Action Fabric applies existing business rules and permissions, so the actions available reflect each organization's own configuration and access.

The rest is a roadmap, and the dates are specific. A planned Codex integration in ServiceNow's AI Workflow Factory, plus OpenAI models powering Build Agent and Autonomous Engineer, is expected in 2027. A joint security solution combining OpenAI's cybersecurity capabilities with ServiceNow's asset, identity, and business context is expected in early access in January 2027, with general availability in the second quarter of 2027. Voice in the Super Intelligence Control Tower is expected in the first quarter of 2027. ServiceNow plans to give all employees access to ChatGPT Work and Codex, and OpenAI joins ServiceNow's Control Tower Alliance as a founding member.

The pattern is worth naming. Last week's roundup covered Anthropic's proactive Duet agent and OpenAI's cheaper computer-using Terra model. This week the agent story moves from the model to the system of record: the ticket queue, the change request, the approval chain. Agents that can read a screen are impressive. Agents that can act inside the workflow where work actually gets approved are the ones enterprises will pay for, and this partnership is a direct bid for that ground.

5. Liquid AI's LFM3: Six Small Models, One Shared Brain, About 1 GB

The week's counterweight to Google's frontier launches came from Liquid AI, which released LFM3 on October 27, a family of six open-weight models in the 2-billion-parameter range, all branching from the same LFM3-1.7B-Base backbone, reports Unite.AI. The family covers text (LFM3-1.7B), vision and video (LFM3-VL-2B), audio (LFM3-Audio-2B), an omni model combining all of them (LFM3-Omni-2B), and two decision models (d1-2B and d1-Omni-2B) that return calibrated answers instead of generated text. The base was pre-trained on 22 trillion tokens, and the text model carries a 256,000-token context window.

The pitch is size, not scale. The text model fits in about 1 GB of memory at 4-bit precision, and Liquid says it decodes 1.8 to 2.8 times faster than the next model across the three on-device setups it tested, with support for phones, a Raspberry Pi 5, laptops, and NPUs. Those are Liquid's own measurements, not independent tests, and the company's own model card says the text model is not recommended for agentic coding or complex multi-step reasoning. Commercial use is free below a $10 million annual revenue threshold under the LFM Open License. The contrast with Carbon is the point: one week, one lab ships the strongest security model it has ever built behind a gated rollout, and another ships six small models anyone can download and run on a phone. The AI industry is splitting into both directions at once, and this week proved it.

How the Week's Stories Stack Up

The thread across all five: AI is moving from demo to deployment, in both directions. At the top end, models are being handed to defenders first and measured on whether they can find and fix real flaws. At the bottom end, models are shrinking onto phones and single boards. In the middle, the cloud bill for all of it is doubling, and enterprises are wiring agents into the workflows where approvals actually happen. Last week was about who AI answers to and what it costs. This week is about where AI runs, who gets it first, and what it can actually do once it arrives.

If you like quick video takes on how new open AI projects work, browse the repo signals for 30-second overviews. For last week's roundup, see This Week in AI: October 28, 2026.

Frequently Asked Questions

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google's frontier model announced on October 28, 2026, built for cybersecurity defense and long-horizon engineering work. It scored 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching, and ranks first on the Artificial Analysis Intelligence Index v5 at 59. Access starts with a subset of cyber defenders through Google's Fairwind Program, at an introductory price of $2 per million input tokens and $10 per million output tokens.

What is Genie 4?

Genie 4 is Google DeepMind's world model unveiled on October 27, 2026, for real-time interactive simulation. It generates a scene from a text or image prompt that users can steer with typed commands while it runs, and it is DeepMind's first real-time world model to generate audio natively, including music, ambient sound, and character dialogue. Scenes stay consistent for about two minutes, and access is currently a research preview for trusted testers, with Google AI Ultra subscribers in Project Genie next.

What did Microsoft and Alphabet report?

On October 28, 2026, Alphabet reported third-quarter revenue of $127.6 billion, up 25%, with Google Cloud up 103% to $30.7 billion, and projected capital expenditures of $195 billion to $205 billion for the year. Microsoft reported fiscal first-quarter revenue of $92.3 billion, up 19%, with Azure revenue of $33.2 billion, up 48%, its first disclosure of Azure revenue in dollars.

What does the OpenAI and ServiceNow partnership do?

Announced on October 28, 2026, the expanded partnership connects ServiceNow workflows with ChatGPT and Codex. A ServiceNow MCP plugin available in the OpenAI marketplace lets teams find similar incidents, assess change risk, and update ServiceNow records from inside ChatGPT, governed by ServiceNow's Action Fabric permissions. A joint security solution is planned for early access in January 2027 and general availability in the second quarter of 2027.

What is Liquid AI's LFM3?

LFM3 is Liquid AI's family of six open-weight models released on October 27, 2026, each around 2 billion parameters and branching from one 1.7-billion-parameter base pre-trained on 22 trillion tokens. It covers text, vision, audio, omni, and decision models, with the text model fitting in about 1 GB at 4-bit precision and a 256,000-token context window. Commercial use is free below $10 million in annual revenue under the LFM Open License.

Conclusion

This week Google shipped at both ends of the spectrum: a gated frontier model for defenders and a playful world model with sound, while Microsoft and Alphabet showed that the cloud underneath it all is growing at triple and near-double digits and costing more than ever to build. OpenAI and ServiceNow moved agents into the workflow, and Liquid AI moved models onto the phone. The models will keep improving. The winners of the next phase will be the ones that put AI where the work already happens, whether that is a security operations center, a ticket queue, or a device in your pocket.

Read this article in Bahasa Indonesia: Berita AI: 30 Oktober 2026.