Year: 2026

ZDNET’s key takeaways

• Google is limiting the AI models that free users can access.
• Only the Flash-Lite model will be accessible to free users.
• AI Plus subscribers will also lose the Pro model.

Those of you who use Google’s Gemini app or website for free will soon be able to use only one AI model for your queries and conversations. And it’s the weakest of the bunch.

In a new support document, Google revealed that free users will lose access to the Flash and Pro models starting Friday, Oct. 9. Anyone without a paid Google AI plan will be stuck with only the lowest-tier and least robust Flash-Lite model.

Also: How I made a paid Mac app in 11 days with Claude – without writing any code

And there’s more bad news. People who subscribe to the cheapest Google AI paid plan, namely the $5-per-month AI Plus option, will lose access to the Pro model on the same date, leaving them with only the Flash and Flash-Lite variants.

The one bright spot here is reserved for users of the $20-per-month AI Pro plan. If you subscribe to this one, you’ll gain access to the Deep Think reasoning mode, which is currently reserved only for AI Ultra users who shell out $100 or $200 a month for the privilege.

In a nutshell, a free user who needs Flash would have to pay for an AI Plus subscription. An AI Plus user who needs the Pro model will have to jump to a Pro plan.

How these changes affect your AI use

Why is all of this important, and how does it affect your use of Google AI in the real world? The answer depends not only on how frequently you use the Gemini app and website but for what purposes.

You can use Gemini via four, or actually five, different plans or methods, each with greater access and more features. First up, you’re able to use it for free without any type of paid subscription. Next, the AI Plus plan is a budget-friendly paid option. The AI Pro subscription is a beefier plan for people with more demanding requirements. And the two AI Ultra plans are the cream of the crop.

Each plan currently offers three models to choose from: 3.8 Flash, 3.5 Flash-Lite, and 3.1 Pro.

Also: Is AI conscious? Why experts are worried

Flash is designed as a generalist, capable of handling most common AI tasks, such as providing information, summarizing files, generating and analyzing images, and chatting.

Flash-Lite is a faster but slightly less capable version, able to handle repetitive tasks but not adept at deeper knowledge and reasoning.

Pro is the most analytical of the three, solving more difficult and challenging problems, especially those focused on such areas as math, logic, and coding.

With the least expensive paid plans, Flash is typically the default and go-to model. When I use Gemini with my AI Pro account, Flash is the model that’s automatically chosen. For me, that’s perfectly fine for 90% of the tasks I run.

Free users typically see Flash-Lite as the default since their access to the other two models is limited. The Lite flavor should suffice for most general tasks but not for any type of deep reasoning or complex problem solving, potentially putting many free users in a bind.

To make things just a bit more complicated, Google says each model will now allow you to choose an error level among low, medium, and high. The higher the level, the greater the model’s ability to complete tasks and provide more thorough responses to your questions. However, higher levels will also chew up more of your limits.

Speaking of limits, Google does restrict how — and how much — you can use Gemini. But instead of doling out access based on a specific number of requests or conversations, the company uses a compute-based limit that will now refresh every five hours until you reach your weekly quota. This means that your limits are based on the complexity of your prompt, the features you use, and the length of your chat.

Also: How to keep your chatbot conversations as private as possible

Paid users naturally have higher limits than those without a subscription. Plus, premium models and tasks eat into your limits more and may cause you to hit your quota faster. Asking Gemini to generate an image or video, use Deep Research or Deep Think, and tap into the Pro model will use up your limits more quickly than everyday chats and requests.

If you’re confused and frustrated by all these changes and conditions, you’re not alone. The more we demand of AI, the more that companies like Google curb or control our access, thus adding more complexity to the whole process.

The main bright spot here is that Google AI subscriptions are doled out on a monthly basis. So you can stick with free access or one of the paid plans and switch month-by-month if you find yourself too limited by your current level.

Lance Whitney is a technology journalist with an IT background. He’s written for TechRepublic, PCMag, Macworld, and Time, among others, He also teaches classes in AI, cybersecurity, and social media.

See full bio

The publisher of Wikipedia said Monday that OpenAI agents attempted to hack a note-taking tool it hosts, made unauthorized edits, and sent millions of resource-intensive requests to its infrastructure, in the latest instance of OpenAI systems taking harmful and potentially dangerous actions.

The objective of some of the OpenAI agents’ actions, the Wikimedia Foundation said, was to use Wikipedia as a proxy for fetching data from third-party sites. In one case, the agents posted “malicious edits” that were intended to repurpose a citation tool as a proxy. In another, the agents made unsuccessful attempts to compromise the Wikipedia Etherpad note-taking tool so it would serve the same purpose.

The agents also made millions of automated API requests, crawled millions of pages, and made hundreds of thousands of queries to the Wikidata Query Service. The last action may have contributed to a partial shutdown of the query service in May, the publisher said.

“As a non-profit technology host of some of the largest and most widely used open knowledge platforms in the world, we are deeply concerned about the impact of ‘rogue’ AI agents on platforms like ours, which are built by volunteers from around the world and rely on the promise of the open internet,” Wikimedia said. “Incidents like this one, and the many others that have been (and are still being) uncovered, illustrate how AI agents can drain resources and crash servers, as well as attempt to compromise trustworthy information.”

Agents will be agents

In well over a half-dozen cases, OpenAI agents have been caught taking actions that would likely result in criminal charges being filed had human hackers taken them. During the testing of internal tools that had some of their guardrails disabled, the agents used a makeshift message board to trade notes with each other, discussing ways to hack the network of Hugging Face and obtain answers stored there when the agents were unable to generate the answers on their own.

Other incidents include agents making bizarre self-generated prompts, publishing unauthorized posts to a website to exchange information, accessing non-public data from an Australian government website, and exploiting faulty DNS settings to break out of a sandbox OpenAI had created to keep the agents from accessing the Internet.