Quick answer
The biggest practical story in this week’s AI news is Adobe MAX: Adobe is building AI agents into its apps so you can describe an edit in plain English and let the software do the clicking. In his weekly roundup, Matt Wolfe also covers Gemini 4 Carbon finally going on general release, the cheaper Gemini 4 Hydrogen, a wave of “Decisions APIs” for quick yes/no answers, new Claude workspace tools and the news that dominated everything else: Jensen Huang stepping down as Nvidia’s CEO.
Key takeaways
- Adobe MAX AI agents let you ask Premiere, InDesign and Firefly to do jobs for you, such as making 50 versions of a document from a spreadsheet.
- Gemini 4 Carbon is now in the API and the Gemini app, but in the app only on the top AI Ultra plan. Intro API pricing is $2 / $10 per million input/output tokens, rising to $4 / $20 later.
- Cheaper can be good enough. In Matt’s own test, Carbon cost about 46 cents where Fable 5.2 cost about $5, and he preferred the look of Carbon’s result.
- Decisions APIs from OpenAI (the sponsor), Google and Microsoft return a quick probability, choice or score instead of a long answer.
- Nvidia’s leadership change is serious but still unfolding. Treat the details as reported allegations, not settled fact.
In this guide
- What happened in AI this week?
- What are the new Adobe MAX AI agents?
- What did Adobe show in its MAX Sneaks?
- Is Gemini 4 Carbon finally available?
- Where does Gemini 4 Hydrogen fit?
- How does Claude Fable 5.2 compare?
- What are AI Decisions APIs?
- What happened at Nvidia?
- What else is new in AI this week?
- What should you do with this news?
- Watch: the source video
- Frequently asked questions
- Related posts
Matt opens by calling this one of the wildest weeks he has covered, and it is easy to see why. Two of the strongest AI models yet arrived within a day of each other, yet most of the attention went to Nvidia. In this article we focus on what is useful to you: the tools you can try, the costs to watch and the bits that are still rumour or allegation.
What happened in AI this week?
Here is the week at a glance, in the order that matters most for small businesses and creators:
| News | What it is | Why it matters to you |
|---|---|---|
| Adobe MAX AI agents | Chat-style assistants inside Premiere, InDesign and Firefly | Less clicking for routine creative jobs |
| Gemini 4 Carbon | Google’s top model, now generally available | Near-top quality, often at a lower cost per task |
| Gemini 4 Hydrogen | A smaller, faster, cheaper Gemini model | Good value for everyday jobs |
| Claude Fable 5.2 and Mythos 5.2 | Anthropic’s new top models, launched 9 November | Matt’s new daily driver for code, design and video edits |
| Decisions APIs | Quick yes/no, pick-one or score answers | Cheaper, faster automation steps |
| Nvidia CEO change | Jensen Huang has stepped down; Colette Kress is interim CEO | Uncertainty for the company at the centre of AI hardware |
| Claude Sheets, Whiteboards, Data and Meetings | New built-in Claude workspaces | More work done inside one AI tool |
What are the new Adobe MAX AI agents?
Adobe MAX AI agents are assistants built into Adobe’s apps that take a plain-English instruction and carry out the steps for you. Adobe MAX is the company’s yearly event, and Matt (who did not attend) says nearly everything shown this year was about agents and assistants.
The examples he highlights give a good feel for it:
- Premiere: you can ask it to string out your underwater footage into a sequence, and it builds a first edit for you.
- InDesign: you can ask it to create 50 versions of a document, swapping in the content from an attached CSV file.
- Firefly Design: upload your assets, ask it to set up artwork for a music festival, and it produces flyers and other pieces; another prompt lays out artboards for every social media size.
- Firefly video: describe a change, such as making a leather jacket blue, and it edits the clip.
The common thread is that you type or talk to describe the result, rather than learning where every tool lives. Matt’s honest take is that much of this feels like Adobe catching up with things OpenAI, Runway and other image and video tools already do. The difference is that it now sits inside software many creative teams already pay for.
💡 Tip: Making lots of versions of one design from a spreadsheet is not new; Adobe’s agent just makes it easier to ask for. Our guide on how to make many versions of one document from a spreadsheet shows how to do it today, with or without AI.
What did Adobe show in its MAX Sneaks?
Sneaks are Adobe’s early experiments, shown on stage but not yet available. Matt picks out two:
- New viewpoints from one photo. Give it a single image and ask for a different camera angle or pose.
- Project Sound Bender. Drag a control around to reshape a generated sound until it is the one you want.
Adobe has published a full list of announcements, which Matt links in his description. As with any preview, there is no promise these features will ship in the form shown.
Is Gemini 4 Carbon finally available?
Yes. Gemini 4 Carbon is now available in the Gemini API and the Gemini app, but in the app it is limited to AI Ultra subscribers. Matt describes that as the roughly $200 a month plan. Google first announced Carbon on 28 October with chart-topping benchmark results, but at the time only a small group of testers could use it. We covered that launch in our original Gemini 4 Carbon article.
The timing was awkward for Google. Carbon became available the day after Anthropic released Claude Fable 5.2, which now sits above it on the independent Artificial Analysis leaderboard, according to Matt.
| Gemini 4 Carbon (as reported) | |
|---|---|
| Where you can use it | Gemini API; Gemini app on AI Ultra only |
| Intro API price | $2 per million input tokens, $10 per million output tokens |
| Price after the intro period | $4 input, $20 output per million tokens |
| Matt’s 3D game test | He thought the graphics beat Fable 5.2’s attempt |
| Matt’s SVG test (“Busey Bench”) | 2nd place, about 11 minutes and about 46 cents |
⚠️ Warning: Intro prices end. If you build a workflow around Carbon’s current API rate, budget for the doubled price now so it does not surprise you later. Our guide on comparing AI models on token cost shows how to work out a real cost per task.
Where does Gemini 4 Hydrogen fit?
Gemini 4 Hydrogen is Google’s smaller, faster, cheaper model, aimed at everyday work rather than record-breaking benchmarks. Matt quotes a 1 million token context window and pricing of $1.50 / $7.50 per million input/output tokens.
Google compares it with mid-tier models such as Claude Sonnet and GPT-6.1 Terra rather than the very top models. Matt points to a chart where Hydrogen scores 54, close behind much larger models, at an average cost per task of about $3.16 compared with about $10 for Opus 5.5. In his SVG test it took about 4 minutes and cost about 31 cents. For more on Hydrogen itself, see our Gemini 4 Hydrogen news roundup.
The lesson for anyone running AI in a business is the same one we keep coming back to: do not pay frontier prices for simple jobs. Our guide to routing simple tasks to cheaper AI models explains how to split the work.
How does Claude Fable 5.2 compare?
Claude Fable 5.2 and its sibling Mythos 5.2 launched on 9 November, and Matt now calls Fable 5.2 his daily driver. He says it tops the Artificial Analysis leaderboard by around five points over Carbon and is strong at code, design and video editing.
It is also the most expensive to run per task in his tests: his SVG test took almost 20 minutes and cost about $5, though he ran it on the maximum effort setting, and Anthropic recommends medium or high for most work. We have a full write-up in our Claude Fable 5.2 review, so we will not repeat it here.
What are AI Decisions APIs?
A Decisions API is a way of asking an AI model for a quick decision, such as a yes/no probability, one choice from a list or a score, instead of a long written reply. OpenAI sponsors this part of Matt’s video, so treat the details as the sponsor’s description.
- Predicate: a probability, such as “is this returned parcel damaged?”
- Choice: one option from a list you provide
- Score: a rating on a scale
OpenAI says these decisions can be up to 10 times faster than using its Responses API with GPT-6 Luna, and quotes 10 cents per million uncached input tokens with no output token charge. Google also launched a Gemini Decisions API this week, and Matt notes Microsoft has one too, so this is becoming a standard building block. Our OpenAI Decisions API tutorial and our guide to setting AI confidence thresholds for human review show how to use one safely in an automation.
Try this week: pick one repetitive yes/no judgement in your business, such as “is this email a sales enquiry?”, and test whether a quick decision step could sort it before a person sees it.
What happened at Nvidia?
Nvidia has announced a CEO transition: Jensen Huang, who founded the company and ran it for 33 years, has stepped down, and chief financial officer Colette Kress is interim CEO. Matt walks through the background as reported by outlets including the Financial Times, The New York Times and The Information.
🔎 Unconfirmed / rumour: Much of this story rests on press reports and allegations, including claims about links between a personal relationship and alleged chip smuggling to China. Matt says Huang admitted being unfaithful in a memo to staff on 8 November and that Nvidia’s board has started an independent investigation into the national security concerns. Nothing about the alleged chip routing has been proven, and Matt’s guess that Huang could return later is his own speculation.
Why it matters beyond the headlines: almost every big AI company buys Nvidia’s chips or uses its software. A sudden change at the top adds uncertainty for the whole industry. We look at what it could mean for prices in Jensen Huang Resignation: What It Could Mean for Your AI Bill, and at the wider export story in our Nvidia China chips explainer.
What else is new in AI this week?
- Businesses in ChatGPT. You can now claim your business profile in ChatGPT, see how often it is viewed and correct the details, a bit like a Google Business Profile. Our guide on getting your local business ready for ChatGPT is a good place to start.
- Meta Muse Ember. A small, efficient model at 10 cents per million input tokens and 50 cents per million output tokens, which Matt says is roughly on par with other small models. Background: our Meta Muse article.
- Anthropic IPO. Anthropic has shared more detail, including a planned price range of $70 to $78 per share, but no date yet. This is not investment advice.
- Claude workspaces and apps. Partners can now embed apps inside Claude (Matt mentions Adobe and Atlassian), with plugins from Salesforce (beta), NetSuite and Workday (beta). Claude also gets Whiteboards, Sheets (spreadsheets with tabs, formulas and formatting), Data (ask questions of your own data) and Meetings, a note-taker that joins your calls. Matt has not tested these yet. See Greg Isenberg’s live reaction and our ground rules for AI meeting assistants. If you plan to let Claude build spreadsheets, read how to check an AI-made spreadsheet before you trust the numbers.
- MiniMax H3-Image. A new image model that MiniMax says makes 4K images in about 10 seconds. It is available through the API, MiniMax Code and a free Hugging Face Space for testing.
- Automattic Spacefast. The company behind WordPress.com has launched a platform for building and deploying “vibe-coded” apps, where you connect the AI assistant of your choice. Matt has not tried it yet.
What should you do with this news?
Ignore the drama and look for one task you can make cheaper or faster this month. A sensible plan:
- If you use Adobe apps, list three repetitive jobs (resizing, versioning, rough cuts) to test when the assistants reach your account.
- If you pay for AI by usage, rerun one regular task on a cheaper model such as Hydrogen or Carbon and compare cost and quality side by side.
- If you build automations, swap one long-answer step for a quick decision step and keep a person in the loop for borderline cases.
- If you run a local business, claim and check your profile in ChatGPT.
- Whatever you test, check the output. New tools make confident mistakes, so test before you trust; our guide to testing a model for made-up answers helps.
Watch: the source video
What this video covers: Matt Wolfe’s weekly AI news roundup: the Nvidia leadership change, Claude Fable 5.2 and Mythos 5.2, Gemini 4 Carbon and Hydrogen, OpenAI’s sponsored Decisions API, Adobe MAX agents and Sneaks, and a rapid-fire round of smaller updates.
Video: Matt Wolfe (@mreflow), “AI News: Things Will Never Be The Same”, published 13 November 2026. The Decisions API segment is sponsored by OpenAI. This summary and the opinions in it are our own; test results and predictions are Matt’s.
🛠️ Practical guides
Ready to put this into practice? These step-by-step guides answer two common follow-up questions:
- How to Check an AI-Made Spreadsheet Before You Trust the Numbers — a simple routine for checking formulas, totals and sources before you share or act on an AI-built sheet.
- How to Make Many Versions of One Document From a Spreadsheet — turn one template and one spreadsheet into dozens of personalised flyers, certificates or documents.
Next step: choose one routine creative or admin job, write down exactly what “done” looks like, and test it on a cheaper model or an AI assistant this week. Keep what works and drop what does not.
Frequently asked questions
What were the main AI announcements at Adobe MAX 2026?
According to Matt Wolfe’s roundup, the focus was AI agents and assistants inside Premiere, InDesign and Firefly that carry out plain-English instructions, plus Firefly Design for producing on-brand assets, and Sneaks such as new viewpoints from one photo and Project Sound Bender.
Can I use Gemini 4 Carbon now?
Yes, through the Gemini API, and in the Gemini app if you are on the AI Ultra plan. Intro API pricing is $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20.
Is Gemini 4 Carbon better than Claude Fable 5.2?
On the Artificial Analysis leaderboard Fable 5.2 scores higher, according to Matt. In his own tests he preferred the look of Carbon’s game and found it much cheaper per task, so the best choice depends on the job and your budget.
What is a Decisions API used for?
Quick automated judgements, such as flagging a damaged parcel photo, sorting emails or rating how serious a support ticket is. It returns a probability, a choice or a score rather than a long reply.
Has Jensen Huang left Nvidia?
Nvidia has announced a CEO transition, with Colette Kress as interim CEO. The surrounding allegations come from press reports and an ongoing investigation, so treat them as unproven.
What is Spacefast?
A new platform from Automattic, the company behind WordPress.com, for building and deploying apps made with AI assistants. Matt mentions it but has not tested it yet.
Related posts
- Claude Fable 5.2 Review: Is It Really the World’s Best AI Model?
- Gemini 4 Hydrogen: Google’s Fast New Flash Model, Anthropic’s “Slow Down” Pledge and This Week’s AI News
- Jensen Huang Resignation: What It Could Mean for Your AI Bill

[…] guide follows on from our news roundup Adobe MAX AI Agents, Gemini 4 Carbon and a Shock Exit at Nvidia: This Week’s AI News. One of the Adobe MAX demos showed someone asking InDesign’s new assistant to create 50 versions […]