It's the first time in the history of the world.
Today, when we went online, the sky was falling.
Countless older users have come up with a reminder: The days of Google's high-end "white-collar" models are over!
Google announced a series of new measures on its official support page.
Starting on October 9, Google will fully adjust access permissions for the Gemini model, incorporating more advanced models into the paid wall.
Free users are downgraded to using only the most basic Flash-Lite model;
AI Plus subscribers will lose the right to use the Pro model and only keep the Flash series;
AI Pro and Ultra Advanced subscribers can continue to use the Pro model.
As a result, many users who build workflows based on Google's low-level packages will be severely disrupted.
The netizens shouted: "You spent money on Plus subscriptions and you woke up becoming a second-class citizen?"
At the same time, Google's Antigravity was hit hard.
The Antigravity platform has announced that Claude Opus 5.5 and Sonnet 5.5 will be available to all paid users, with Opus 4.6 and Sonnet 4.6 coming out on November 2.
Why is Google, the "philanthropist" that used to scour the development circle with free Pro quotas, so crazy about cutting packages?
All the clues point to the next-generation Gemini 4 Argon!
It looks like Gemini 4 Argon is about to open.
Let's see how the official support page is updated this time, which is breaking the walls of all web developers.
This time, Google sent a final message to all Gemini users: "Free days are over, you have to pay for top-notch AI".
In this change, the personal user rights of Gemini Apps are divided into four categories.
First of all, yes.Free users without AI subscriptions(In force from 9 October)
The Flash and even the Pro models, which you can occasionally see in white, have all evaporated! The future is left to the free White Panthers, with only the most basic "Flash-Lite" left.
For them, the free version was formally reduced to a lightweight "advanced search box", complex logical inferences, deep long-text analysis, and high-level code debugging.
Secondly, yes.Subscribe to AI PlusThis is the largest wave of protests in the country.
Many users spent money subscribing to Plus, which was to see the high-frequency calling Pro model. However, according to the official announcement, Plus users were denied access to the Pro model, with the exception of Flash-Lite and Flash!
It's too hot to spend real gold and silver and be directly downgraded to free user treatment.
And finally, yes.Honored members of AI Pro and AI Ultra.
Only these two high-end subscribers will retain full access to the Flash-Lite, Flash and Pro models.
And Google's core black technology, Deep Think, has the greatest parallel reasoning ability, and it's limited to both Pro and Ultra users.
Google has officially launched a limitation system based on actual computations, in addition to the limitation.
From now on, models will no longer be "calculated by multiples" and will be replaced by a more rigorous "dynamic consumption system".
Specifically, the system is dynamically refreshed five hours at a time, while overlapping with weekly hardness limits.
The system will charge you a comprehensive fee based on the complexity of your questions, the length of context, the additional features you use (e.g. image/video/music generation, deep research, extended thinking).
Officials also added a new "Effort" slider, divided into low, medium and high. If you want a more detailed answer, you can choose a high degree of effort, but the amount will also evaporate quickly.
When the combination of the two came to light, Google's goal was clear: to compress the cost of computing power, to harvest high net worth users.
But is Google really just to save money?
If you look at it this afternoon, you'll know that it's not that simple because they're creating a huge amount of computational space.
The anti-gravity platform has also begun a massive exchange of blood against ordinary users.
Google has officially announced that Claude Opus 5.5 and Sonnet 5.5 are now fully open to Google AI Pro and Ultra users on the platform!
Google AI Pro is $19.99 a month, Ultra 5x is $99.99 a month, Ultra 20x is $199.99 a month.
The most powerful force of Anthropic is that it is plugged into Google's core toolbox and made available to top paid members.
And the old third-party model is going to get rid of the whole line.
GPT-OSS 120B, which was once integrated into the platform, and the older Claude 4.6, will be officially launched on November 2nd!
At that point, the current third-party model access will be completely ineffective in the existing plan.
Varun Mohan, the head of Google Antigravity, said:
Some have noticed that we have added Claude Opus 5.5 and Sonnet 5.5 to all paid users of Antigravity.
At the same time, we are also very excited about the upcoming Gemini 4 Argon, which will be launched more widely!
It's all over!
If you put it all together, you can understand Google's strategic intentions.
Cutting off the privileges of Pro and Plus users, recycling a huge amount of everyday noise computing; flushing out third-party resource models such as GPT-OSS 120B; and stabilizing the most provocative Pro/Ultra high-end paid users with Claude 5.5.
All of this is for the super giant Gemini 4 Argon!
This also shows a clear signal that the products chosen by users are becoming coding smart objects, not models.
Google is willing to put Claude 5.5 into its own tool because developers want editors, smart circuits, and plugins, and want to switch underlying models without replacing tools.
Antigravity's recent update also points in the same direction with the release in late September of a new Customizations tag page for installing plugins, as well as the ability to send messages directly to sub-smartphones.
The Giant Breaks Out:Gemini 4 Argon is coming?
Now, Google has cut out the Gemini 3.5 Pro and directly moved the version number to 4.0, code named "Argon".
As early as September 30, the Gemini 4 Argon was released, but Google only made it available to top cybersecurity collaborators like Fairwind.
The following hard indicators can be seen from the few test reports that are currently circulating:
1.1M "Output" Token
Note that this is not the context input of 1M, but the entire output generating capacity of 1 million!
Currently, the industry's top model, the output token, is generally limited to between 4k and 64k.
That means you throw a need at it, that it can rewrite the underlying architecture of a medium-sized software system in a single interaction, generate millions of words, or rebuild the underlying structure of tens of thousands of lines of cybersecurity code.
This is why it is preferred to be thrown into the field of cybersecurity in secret test tubes in automated vulnerability mining and defense deployments, where Argon's destructiveness and productivity have reached paramilitary levels.
2. The price of slaughter
Usually this "nuclear-grade" model, the cost of the call, is priceless.
But internal information shows: The Gemini 4 Argon is priced at a staggering $2 / $10 per million (input/output token) after its early launch of the paid API and AI Ultra!
With the scale effect of its self-developed TPU computing cluster, Google is attempting to use cliff-like cost barriers to challenge rivals OpenAI and Anthropic.
All countdowns point to the same time slot on November 2.
This day may be the date of the more widespread launch of Gemini 4 Argon.
The free version has been scrapped; the Plus version has been downgraded, not because Google is getting pissed off, but to support the 4.0 Argon monster.
The free lunch era of AI is over.Obviously, the "honeymoon subsidy period" since the birth of generative AI is over.
Looking back over the past two years, families have been burning billions of dollars' worth of gold and silver for a circle: free flagship models, declining API prices...
But now, computational reasoning has skyrocketed exponentially, and the traditional per-capita SaaS model is collapsing. In the future, accounting fees and calculation will become mainstream.
If you're a lightweight Google user, you can use Flash-Lite without the use of blind upgrades.
If you're a heavyweight user, you can either upgrade directly to Google AI Pro/Ultra, along with Claude 5.5 on the whiteboard Antigravity, or go directly to API calling, for a weight-to-weight ratio.
By the way, look forward to what Google is going to do on November 2nd.