Google has so far allowed free access to its latest Gemini Flash models for users on the free plan. But that will change starting later this week, with the company limiting access to its latest and most powerful models to its paying customers.

An updated support page from Google highlights the upcoming changes to Gemini's model access and limits. Personal account holders with no paid subscription will only have access to Gemini's Flash-Lite model beginning October 9. They will not be able to use the Flash or Pro models.

Even users on the paid plan have not been spared. AI Plus subscribers, Google's cheapest AI plan, will lose access to the Pro model starting October 9.

Note that these changes affect access to Gemini model families. Each family can include multiple versions, and access to a family does not necessarily include its latest model. For example, AI Plus subscribers currently only have access to Gemini Flash 3.6, even though Flash 3.8 is the latest model.

Only AI Pro and AI Ultra plan subscribers will have access to different Gemini models. In fact, AI Pro subscribers will now have access to the Deep Think option, enabling "maximum parallel reasoning," a feature previously reserved for AI Ultra subscribers.

Like Claude and ChatGPT, Google will also introduce a model thinking effort level for Gemini. Irrespective of the model, you can choose from three effort levels: low, medium, or high.

Claude and ChatGPT also have model usage restrictions on free plans, so it's not surprising to see Google follow suit.

This is a big change to Gemini’s model access. It comes within months of Google switching to a compute-used usage model, rolling out a 5-hour and weekly usage limit. Before that, Gemini's usage was calculated based on the number of prompts.

Same usage limits, fewer model choices

The plans’ usage limits remain unchanged. AI Plus subscribers get twice the standard usage limit, while AI Pro subscribers get four times that limit. AI Ultra subscribers get five or 20 times the standard limit, depending on their plan.

Google is clearly making these changes to free up resources for the public launch of Gemini 4 Argon, its most powerful and capable AI model yet. The company benchmarks indicating it beats the latest ChatGPT and Claude models in agentic coding, reasoning, math, and other tasks.

For now, Gemini 4 Argon is only available to trusted cyber defenders through Google's Fairwind Program. But a public release should be coming soon and could happen within the next few weeks.