AI News Roundup – October 29, 2026

TL;DR

• Google announced Gemini 4 Carbon, its strongest model. It ranks #1 on the Artificial Analysis Intelligence Index, but only a subset of cyber defenders get it first.
• Alphabet says the Gemini app hit 1 billion users and Google Cloud revenue rose 103%. Microsoft’s Azure brought in $33.2 billion, up 48%.
• Mistral released the open weights of Mistral Large 4, a 1-trillion-parameter model, at $0.70 / $2.10 per million tokens on its API.
• OpenAI rewrote its Model Spec around five values and retired most of its detailed style rules.
• OpenAI and ServiceNow expanded their partnership. A ServiceNow plugin is in ChatGPT today.
• Anthropic launched smart reports, in-app surveys and pooled budgets so Claude Enterprise customers can see what their AI spend buys.
• GitHub launched Copilot Fusion, and its Octoverse report says agents now start more than 40% of pull requests.

Today’s take: The newest frontier AI is being rationed to defenders and big customers first, while open and auto-routed models make “good enough” AI cheaper for everyone else.

---

What is Gemini 4 Carbon, and who can use it?

Short answer: Google’s new top model, built for cybersecurity and long-running tasks. Right now only a subset of cyber defenders can use it. Google says it will add more trusted partners through its Fairwind Program, then open it to developers, businesses and consumers “as soon as possible.” Google is also taking part in the U.S. government’s voluntary pre-release testing.

Gemini 4 Carbon	Google’s figure
Artificial Analysis Intelligence Index (v5.0)	#1, score 59
SEC-bench Pro (finding vulnerabilities)	88.2%
ExploitGym (exploitation)	54.8%
CWE-bench (patching)	71.0%
WebDev Arena	#4 (1768), up 89 Elo
Introductory price	$2 input / $10 output per million tokens (cached input 95% off)

Google says its teams have seen Carbon work for days on hard engineering tasks, and that it cut unsupported claims by about 2x versus Argon. Carbon arrives just two weeks after Gemini 4 Argon. Benchmark numbers are Google’s own.

Why it matters: The best model now goes to defenders before the public. If you use Gemini today, you’re on Argon, but Carbon’s price is already set for when it opens. Compare costs in Gemini pricing explained, see our Gemini guide, or browse AI tools for security engineers.

Sources: Google · 9to5Google · The Next Web

---

How big is AI demand right now?

Short answer: Bigger than the cloud companies can supply. Alphabet and Microsoft both reported on Wednesday.

Metric (latest quarter)	Figure	Source
Gemini app users	1 billion	Alphabet
Google Cloud revenue	$30.7B, up 103%	Alphabet
Google model API usage	Nearly 32 billion tokens per minute, 4x a year ago	Alphabet
Google Cloud backlog	About $560B	CNBC
Azure revenue	$33.2B, up 48%	Microsoft via GeekWire

Sundar Pichai called the Gemini app “the fastest-growing product in Google’s history.” Microsoft CFO Amy Hood said “customer demand continues to exceed available supply,” and Microsoft disclosed Azure revenue in dollars for the first time. CNBC notes Google Search growth slowed to 12%.

Why it matters: Demand for AI still outruns supply. That’s a big reason new models launch in stages, and why rate limits and prices stay sticky. Compare the big assistants in ChatGPT vs Gemini vs Claude, or see Microsoft Copilot.

Sources: Google (CEO remarks) · Alphabet earnings release (SEC) · CNBC · GeekWire

---

Can you download Mistral Large 4 now?

Short answer: Yes. Mistral released the weights of its improved Mistral Large 4 (“Le Chonk”) on Hugging Face. It’s a 1-trillion-parameter, natively multimodal model trained from scratch in Europe.

Mistral Large 4 vs Gemini 4 Carbon	Mistral Large 4	Gemini 4 Carbon
Weights	Open (download and self-host)	Closed
Input price, per million tokens	$0.70	$2 (introductory)
Output price, per million tokens	$2.10	$10 (introductory)
Who can use it today	Anyone	A subset of cyber defenders

Mistral says it’s the best open-weight model for coding: 48% on the Artificial Analysis Coding Agent Index, tied with GLM-5.3 while using 47% fewer tokens, and 41% on Terminal-Bench 4.0. It has Low, Medium and High reasoning modes. The license is free to use, modify and sell, except that Model-as-a-Service businesses need a separate commercial license. It’s also on OpenRouter, Microsoft Foundry and NVIDIA NIM. Benchmark claims are Mistral’s.

Why it matters: A frontier-class model you can run on your own servers, at a fraction of closed-model prices. That’s useful if your data can’t leave your region. Read what open-source AI means, compare models on our AI models page, or try Mistral’s coding agent, Mistral Vibe.

Sources: Mistral · Hugging Face model card · Hugging Face blog

---

What changed in OpenAI’s Model Spec?

Short answer: OpenAI reorganized the rulebook its models are trained toward around five values: faithfulness to human intent, minimizing harm, honesty and transparency, human agency and care.

• New: a “consider context carefully” section on how conversations and memories should shape what the model thinks you want, plus defaults on supporting human agency and following social norms.
• Retired: most detailed style rules. OpenAI says style should adapt to what people prefer. Defaults like approaching questions objectively stay.
• Renamed: the Spec now calls the model “GPT” instead of “the assistant.”

OpenAI says its models don’t yet fully reflect the Spec.

Why it matters: This is the rulebook behind ChatGPT. Expect it to follow your stated preferences more and lean more on memory and context. Background: what AI safety alignment is.

Sources: OpenAI · Model Spec (Oct 29, 2026)

---

What can ChatGPT do with ServiceNow now?

Short answer: A ServiceNow plugin is available in the OpenAI marketplace today. IT teams can find similar incidents, assess the risk of a change and update a ServiceNow record without leaving ChatGPT. ServiceNow’s Action Fabric applies each company’s existing permissions.

Coming next	Timing (OpenAI)
Enhanced plugin capabilities	By end of 2026
Voice in ServiceNow’s Super Intelligence Control Tower	Q1 2027
Joint security solution	Early access January 2027, GA Q2 2027
Codex in ServiceNow AI Workflow Factory	2027

ServiceNow also plans to give all employees ChatGPT Work and Codex.

Why it matters: Work systems are plugging straight into ChatGPT through MCP. If your company runs on ServiceNow, ChatGPT can now act on tickets with the access you already have. Learn what the Model Context Protocol is, or see ChatGPT Enterprise pricing and OpenAI Codex.

Sources: OpenAI · ServiceNow

---

How can companies see what their Claude spend buys?

Short answer: Three new Claude Enterprise features, generally available now:

• Smart reports: Claude analyzes a sample of a team’s transcripts, groups the work into workstreams, and shows the spend, deliverables and friction points for each, plus recommendations like workflows to turn into skills. Anthropic says reports aren’t for judging individual performance or employment decisions.
• In-app surveys: short questions inside Claude, such as what would break if the team lost access tomorrow.
• Pooled group budgets: one shared monthly budget per team, with individual limits still in place and alerts at 50%, 75%, 95% and 100%.

Availability varies for organizations with HIPAA setups or customer-managed encryption keys.

Why it matters: The question at work is moving from “can we use AI?” to “is it worth it?” Expect your company to start tracking AI spend per team. See Claude Enterprise pricing, our Claude guide, and AI budget caps.

Sources: Anthropic · TechCrunch

---

Do AI agents really start 40% of GitHub pull requests?

Short answer: Yes, by GitHub’s count. Its annual Octoverse report says that by September 2026, more than 40% of pull requests started with an agent’s first draft, up from 17% in March. That was 110 million in September alone. GitHub says its count is a floor, because it can’t detect every agent.

Octoverse 2026	Figure
Pull requests started by an agent (Sept)	40%+ (17% in March)
Agent PRs merged within two weeks (Sept)	90% vs ~86% for others
Copilot users using more than one model (Sept)	61%
Code pushes, year to Sept 30	17.9 billion

GitHub also made Copilot Fusion generally available. Instead of picking a model, you pick an effort level (Standard, High or Max) and Copilot routes each task to one or more models. GitHub says Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the cost per task.

Why it matters: Choosing a model per task is becoming automatic, and agents now write the first draft of a big share of code. Reviewing AI work matters more than typing speed. See GitHub Copilot, Copilot pricing, what an AI model router is, and Cursor vs GitHub Copilot.

Sources: GitHub Changelog · GitHub Octoverse · The Next Web

---

What should you watch next?

• Carbon access: When Google opens Gemini 4 Carbon to developers and consumers, and whether the $2 / $10 introductory price holds.
• Open vs closed: Independent benchmarks of Mistral Large 4 against closed flagships.
• AI at work: Whether more companies measure AI spend per team, and how fast agent-written code keeps growing. Amazon reports earnings today, including AWS.

AI Tools Optimizer is for education only. Figures come from company announcements and the press reports linked above; benchmark claims are company-reported. Verify primary sources before making decisions. This is not investment or legal advice.