noticias ia para empresas gemini pymes

Google has unveiled Gemini 4 Carbon, which tops the Artificial Analysis rankings, but for now it's only going to cybersecurity teams, and its launch price will double once the promotion ends. Spain's central bank confirms that AI use among small businesses has doubled to 30-35%, but only those who use it regularly notice productivity gains. And AI is starting to show up in the financials: Santander puts the value AI has delivered at 194 million euros through September, while Microsoft now rents out the machinery to build your own agents. Catch up on Wednesday's radar (Oct. 28).

Gemini 4 Carbon leads, but you can't use it yet: budget for the final price, not the introductory one

Google DeepMind has unveiled Gemini 4 Carbon, two weeks after Gemini 4 Argon. It is number one on the Artificial Analysis intelligence index with 59 points and stands out in cybersecurity: 88.2% on SEC-bench Pro, which measures finding vulnerabilities, and 71.0% on CWE-bench, which measures fixing them. Google says that, inside its own teams, the model has run for days at a time on long engineering tasks, such as tracking down and fixing subtle bugs in large codebases. But it isn't open yet: this week it goes to a group of cybersecurity defenders, followed by trusted partners in its Fairwind program, and it will reach developers, companies, and consumers "as soon as possible," with no set date. The launch price matches Argon: $2 per million input tokens and $10 per million output tokens; when that introductory period ends, it rises to $4 and $20. What it means for you: don't change your plans today over a model you can't buy yet. Do, however, prepare the 20 real tasks you will use to test any new model the day it opens: that's what lets you make the call in one afternoon. And if you're running the numbers on automating a process, use the full price, which is double: if the business case only works at the launch price, it doesn't work. Source

Bank of Spain: AI now reaches a third of small businesses, but it only pays off for weekly users

Spain's central bank has published an analysis based on its business activity survey, which asked about AI at the end of 2024 and again in early 2026. In two years, usage among SMEs has almost doubled, to roughly 30% to 35% depending on the technology; among large companies it's around 70%, and the gap between them remains as large as in 2024. Almost nine out of ten companies using AI see no effects on employment, but more than a quarter notice productivity improvements. But the number that matters is this one: among those who use generative AI moderately, about 40% see productivity improvements; among those who use it sporadically, only 14%. And looking three years ahead, 57% of large companies plan to reshuffle who does what, compared to 21% of SMEs. What it means for you: having the tool doesn't move you up a step; using it every week in the same process does. Pick a recurring job, such as answering quote requests or preparing reports, decide who now does it with AI, and what that person stops doing. Going from occasional use to habitual use in a specific process is the leap from MANUAL to ASSISTED, and it's exactly the one most SMEs haven't made yet. Source

Santander puts a figure on its AI: 194 million euros through September, calling it just the "tip of the iceberg"

Banco Santander earned 11.176 billion euros through September excluding one-off items, up 16.1%, and 12.405 billion including the capital gain from the sale of its business in Poland. Presenting the results, CEO Héctor Grisi put a number on AI: almost 200 million euros in savings and efficiencies—specifically 194 million in business value through September—calling it the "tip of the iceberg" for a deployment with "many more opportunities." Grisi denied the group is "focused on reducing headcount," although in twelve months it has 4,495 fewer employees (down to 186,151) and 481 fewer branches; the bank handles departures through a voluntary early-retirement agreement with its unions. What it means for you: the point isn't the size of the number; it's that the bank measures its AI in euros. That 194 million is just over 1% of what it spent on personnel and administration in those nine months, so the bulk is still to come, and it will come from redesigning processes, not from handing out licenses. Do the same at your own scale: for every AI use case, log the hours it saves each month and multiply them by the hourly cost. If you can't put a number on it, you don't yet know whether it's working. Source

Microsoft rents out the machinery for its agents: what sets you apart is your written procedure

Microsoft has launched a public preview of the Copilot harness within Foundry Agent Service, its platform for building agents. It's the scaffolding that makes an agent run: Microsoft supplies the agent loop, the sandbox it works in, and context management, while the company provides its knowledge, its procedures packaged as reusable skills, and its tools, connected with explicit permissions and approvals. Separately, the feature that resumes agents running for hours or days after an interruption is now generally available; Microsoft warns that each company's workflow must prevent actions from being repeated when they resume, such as creating the same report twice. The harness is free until December 1, 2026; after that, the execution environment will cost $0.0994 per virtual processor hour and $0.0118 per gigabyte of memory per hour, on top of the AI model itself, which is always billed. What it means for you: the technical side of an agent is becoming something you rent; nobody can rent you how your business works. Write a one-page procedure for the process you want to automate: steps, data, exceptions, and what a person always signs off on. If it doesn't fit on one page, no agent will do it well. And ask whoever builds it: what happens if the agent stops halfway? Does the invoice go out twice? Source

What to watch next

Microsoft and Alphabet reported their results after Wednesday's Wall Street close: the numbers to watch are how fast Azure and Google Cloud are growing and how much they're spending on data centers, because that's what drives the prices and limits you'll pay. And watch for when Google opens Carbon to businesses.