Service update: mitigating abuse and prioritizing traffic #22970
Replies: 342 comments 416 replies
-
|
these changes may require some planning = Switch to Codex btw 🤷🏻♂️ |
Beta Was this translation helpful? Give feedback.
-
|
Had to use both extensively and currently my ranking is Claude Opus 4.6 > Codex 5.3 > Chat 5.4. 5.4 prompts too much for permission despite specific instructions to avoid asking for permission. Codex 5.3 can run for hours but if it encounters something it does not understand or can't grok, it will not figure it out no matter how minute you get in your prompting. Certain long-range tasks, Codex will clobber Claude on, especially scalability, security, or tasks that are more research-oriented, but if there's an issue, nobody is better at getting to the bottom of it than Claude. |
Beta Was this translation helpful? Give feedback.
-
|
this has to be the most idiotic move ever |
Beta Was this translation helpful? Give feedback.
-
If Google's model was better it would be them, and those will be your "secret police" after the election. Grow up. |
Beta Was this translation helpful? Give feedback.
-
|
OK, people at Google, I can understand about the people trying to use the CLI for third-party software. If they would know that CRUSH and OpenCode existed, they certainly would not use their CLI. My complaint is, even for pro users, it is absolute shit. 3.1 pro models and the 2.5 are greatly depreciated. They are shit now. Do you understand? And it's not normal that you get so many less tokens you're gonna lose clients like this. Do you understand? It is just dumb sometimes. You need to work on the fucking road models who gives a fucking Gemini pro fucking preview I was fine with 2.5 when I was working now it's just shit you understand and for three or four days I was not able to work with fucking CLI this is not normal for two weeks you're doing this are you fucking idiots you're gonna lose everything that you get if this continues a couple more days I'm switching to codex codex is amazing chat GPT sucks but codex with the harness with the latest 5-4 or the 5-4 codex it's the top Claude is also acting with fucking predatory tactics to know guys that you can work with free models and free CLIs use open code use crush use aether there are so many there are so many fucking Chinese models now they are fucking amazing if these assholes continue we will all switch to open source I've said it |
Beta Was this translation helpful? Give feedback.
-
|
Appreciate the transparency—changes like these are understandable, but they quietly reshape who gets to build and experiment. Hoping access remains inclusive, because innovation often starts on the free tier. |
Beta Was this translation helpful? Give feedback.
-
|
Is opencode via https://github.com/jenslys/opencode-gemini-auth usage safe? (Pro account) |
Beta Was this translation helpful? Give feedback.
-
Didn't realize OpenCode was hiring. I'm gonna guess the benefits package isn't quite google level. |
Beta Was this translation helpful? Give feedback.
-
|
La verdad, gemini cli me ha funcionado y he crecido en esto de la ia desde el principio, y todavía me sigue funcionando, y los modelos nuevos de google van espectacular, es esperable que los modelos pro dejen de ser gratuitos, saludos desde Colombia |
Beta Was this translation helpful? Give feedback.
-
They get much more information when it's from their own tool. For example, they can A/B test system prompts, test different versions of their edit tool to see which is most consistent, pull more information about the environment to better understand where and why their agent crashes so much, ... Not to mention, if you run out of quota in Opencode, you look for the most cost-effective way to get more usage. That might be Codex or Kimi or Chutes or whatever. If you run out in a Google product, the happy path is to just give Google more money. And there's customer loyalty too - they want you in the ecosystem. These factors means Google gets more value when you use their tool and can better justify running inference at a loss. Obviously Opencode is better. Google isn't saying you can't use Opencode. They just don't want to lose too much money on Opencode users. Why doesn't Codex care? Because sama knows how much y'all love Opencode and the developer sentiment is worth it to them. |
Beta Was this translation helpful? Give feedback.
-
|
@mitchell-wallace if the argument applies to sama, why doesn't it apply to google? google has no interest in developers reputation/karma? |
Beta Was this translation helpful? Give feedback.
-
|
It may work and feel convenient, but tools like that rely on unofficial authentication flows, which brings a few risks like security and policy issues.Its okay for learning/testing, but avoid for production or anything sensitive. |
Beta Was this translation helpful? Give feedback.
-
|
L move.
|
Beta Was this translation helpful? Give feedback.
-
|
Starting March 25, 2026, Gemini Pro models will only be accessible via paid subscriptions. Free tier users will be limited to Gemini Flash models. Upgrade to a Pro or Ultra plan to access Gemini Pro models. But where does that leave paying PLUS users. They are neither free or PRO/ULTRA |
Beta Was this translation helpful? Give feedback.
-
|
I have that same question, because the wording is unclear. If I'm on the 2TB google one premium plan does that mean I will or will not have access to gemini pro models via Gemini CLI on the 25th? If not, does the 2TB "Google AI Pro" tier include it? |
Beta Was this translation helpful? Give feedback.
-
https://one.google.com/about/google-ai-plans/#compare-plans According to the Google One AI Plans page, Gemini Code Assist and Gemini CLI are listed as "Limited", so the exact available capacity is unknown. |
Beta Was this translation helpful? Give feedback.
-
|
I didn't know AI plus subscription is free, refund me the charges then, it's peanuts to you anyway. |
Beta Was this translation helpful? Give feedback.
-
|
They just increased the price for Workspace because it now offers such a huge benefit through Gemini amongst other things. Then around a month later they remove the ability to use all except flash models. It's time to de-Google my infrastructure. |
Beta Was this translation helpful? Give feedback.
-
|
This post doesn't mention the decrease in flash quota , these days it seems it's not 1000 free requests anymore , more like 250 or so , is this correct? And is it on purpose? If so the readme needs to be updated as well |
Beta Was this translation helpful? Give feedback.
-
|
The readme doesn't mention anything about Flash quota. |
Beta Was this translation helpful? Give feedback.
-
|
It says 1000 free requests what is the intended meaning then, it definitely makes you think it's flash especially since that was actually the case, it's just these days it got nerfed |
Beta Was this translation helpful? Give feedback.
-
|
Its 150 requests per day with the flash models |
Beta Was this translation helpful? Give feedback.
-
|
Good question. Our published limits include totals across model calls including Pro (when applicable), Flash and Flash lite. |
Beta Was this translation helpful? Give feedback.
-
|
In terms of purchasing, is AI Pro/Ultra the only option? How about Gemini Code Assist for enterprises? Or organizations that purchase Gemini Enterprise? Can you please confirm? |
Beta Was this translation helpful? Give feedback.
-
|
Good question. Gemini Code Assist for Enterprise remains an option. |
Beta Was this translation helpful? Give feedback.
-
|
But it seems to be just as bad about waiting and never connecting to an AI. Basically the entirety of Google AI services are down now or extremely lagged to the point of being unusable. |
Beta Was this translation helpful? Give feedback.
-
|
Gemini Code Assistant Standard と Gemini Code Assistant Enterprise は 区別されていますか？ どちらも有料のプランです。 |
Beta Was this translation helpful? Give feedback.
-
|
I am using Gemini CLI through its official ACP protocol for this Same interface Zed and JetBrains use. Wanted to flag this in case there's anything to keep in mind for automated/daemon usage of the CLI regarding TOS and fair use. Would love your thoughts and feedback on this as we want to prevent any abuse issues. |
Beta Was this translation helpful? Give feedback.
-
|
I could be misunderstanding so please reply back with more information, if you think it would be helpful. However, based on what you're describing, that sounds like a legitimate use of the product. |
Beta Was this translation helpful? Give feedback.
-
|
Please clarify the allowed usage through the Gemini CLI's Agent Client Protocol (ACP) implementation Both of these are intended to enable third-party tool integration with Gemini CLI to route requests to Gemini CLI and both support leveraging the OAuth subscription configured inside Gemini CLI natively. There are people building solutions based on these like @SaschaHeyer mentioned. Are you changing them? Are they going away? Are you removing OAuth subscription from them? The community needs to know before investing time building more integrations. Thank you. |
Beta Was this translation helpful? Give feedback.
-
|
I need to know as well. I've built a tool that automatically creates sandbox containers for projects where the Claude Code, Gemini CLI, and Codex settings, plugins, skills, MCP servers, and OAuth are all inherited. So the users are still using Gemini CLI exactly as normal but within a more secure sandboxed container. I doubt this is a violation of any of Google's policies since it's basically a symlink. But it'd be nice to know for sure. |
Beta Was this translation helpful? Give feedback.
-
|
@bdmorgan @ryanjsalva Please can we get a clarification to a legitimate question that impacts users and devs? Thank you. What use cases are allowed via your A2A and ACP implementations that are designed to allow integration with third-party tools and support using the user's OAuth-authenticated session that already exists in Gemini CLI. |
Beta Was this translation helpful? Give feedback.
-
|
+1 I'd like to ensure that both:
... will never violate the ToS mentioned here. AFAICS these don't bypass the OAuth procedure used by Gemini CLI and it extends how Gemini CLI can be used. You guys already support headless mode for automations/programming scenarios. However, the words here, "using Gemini CLI oAuth with third-party software," are super vague and spread uncertainty. As mentioned in another reply, even using Gemini CLI with a terminal app can fall into this category. |
Beta Was this translation helpful? Give feedback.
-
This is very vague can you provide more details which licenses get the higher priority and what defines account standing? |
Beta Was this translation helpful? Give feedback.
-
|
What bizarre moves. By capping limits for users, especially Pro ones, you’re preventing them from building API-based apps that would ACTUALLY bring you money and scale to end-customers. I’m sure you’ve run the numbers, but awareness among potential future users is incredibly low. I run AI training sessions, and when I showcase Gemini-CLI or Antigravity, people are absolutely floored because they only associate AI with ChatGPT. Obviously, after these nonsensical limit restrictions for PRO (and I’m not talking about usage outside of CLI or Antigravity), I’m done promoting you. I was doing it for free, and now, on top of that, these limits make it impossible for me to use the tool normally or build anything worthwhile. These are very poor moves that undermine many of your indirect revenue streams. |
Beta Was this translation helpful? Give feedback.
-
|
seems like they aren't good in systemic thinking. weird. |
Beta Was this translation helpful? Give feedback.
-
I thought they had those big thinking models. |
Beta Was this translation helpful? Give feedback.
-
|
I used to think Claude was too expensive, but now, with Google’s restrictions, Antigravity is even more expensive than Claude. That makes no sense, so I might as well just stick with Claude. |
Beta Was this translation helpful? Give feedback.
-
|
Why not just explicitly block / ban users with openclaw? After reading the doc that seems to be the only ISSUE. |
Beta Was this translation helpful? Give feedback.
-
|
Yeah, I think the problem is detecting when it's used by OpenClaw. |
Beta Was this translation helpful? Give feedback.
-
|
Why?? If I pay for credits, allow me to use them as I want |
Beta Was this translation helpful? Give feedback.
-
|
Ofc credits should be allowed. The ones with problem is OAuth, as I think so. |
Beta Was this translation helpful? Give feedback.
-
This does not work? I can't seem to be able to create support ticket to try to restore access to gemini-cli (with Pro subscription), Is there any option I can get some support or at least clear answer so I can cancel, (my case)? @ryanjsalva Btw the link should be |
Beta Was this translation helpful? Give feedback.
-
|
a really bold and wrong move, i guess i'll start looking for alternatives |
Beta Was this translation helpful? Give feedback.
-
|
i am using now Qwen Coder CLI too with Gemini CLI, both as a Consultant (Gem) and the Developer (Qwen), i also will search for another alternatives for Gemini CLI to work with Qwen, |
Beta Was this translation helpful? Give feedback.
-
|
Agreed for moving to qwen. Qwen as light model, Claude ad heavy ones. |
Beta Was this translation helpful? Give feedback.
-
|
@AhmadMWaddah KiloCode has always free tier LLM, and it can be wired to QwenCode, which is one of the best alternatives at the moment. |
Beta Was this translation helpful? Give feedback.
-
|
lol, lmao even, yep that's over. just ban openclaw users.
|
Beta Was this translation helpful? Give feedback.
-
|
Why?? If I pay for credits, allow me to use them as I want |
Beta Was this translation helpful? Give feedback.
-
Yeah it doesn't really make sense isn't that what the quotas are for lol |
Beta Was this translation helpful? Give feedback.
-
|
LOL. |
Beta Was this translation helpful? Give feedback.
-
|
Now the CLI is useless with flash models for free use. It's a shame... although blocking the lobster lovers would have been enough.🤌🏻 |
Beta Was this translation helpful? Give feedback.
-
|
I agree. Flash models should not be anywhere near a terminal. |
Beta Was this translation helpful? Give feedback.
-
|
Well, after mid-March, gemini cli just feels "dumber" than before. And slower. Either it is routing, or restrictions, hard to reverse-engineer. Pre mid-March I've been doing pretty amazing stuff with gemini-cli and have been able to stay within 1000 RPD quota most of the time. Now quota is burning in few hours with much worse outcomes :-(. |
Beta Was this translation helpful? Give feedback.
-
|
With OpenCode and Superpowers, it’s working fine for me, even with Gemini 3 Flash Preview. I think it can also be used with gemini-cli, but gemini-cli often removes lines of code while generating them. That’s the main issue with gemini-cli. |
Beta Was this translation helpful? Give feedback.
-
|
China AI opensource disruption of US AI bubble seems to be working ;-) |
Beta Was this translation helpful? Give feedback.
-
What the hell is wrong with you Google? |
Beta Was this translation helpful? Give feedback.
-
|
Don't forget, Google can review what prompts you are sending to the Gemini CLI or Antigravity; they just don't use those prompts for training. They can even detect whether a prompt is triggered by an agent or a human by looking at the prompt style and signature. If you review the prompts sent by an agent (using the API and enabling prompt logging), you will see that agents are burning token services. How can someone pay $20 or $250 and expect to get $1,000 in benefits in return? |
Beta Was this translation helpful? Give feedback.
-
The issue is you don't have any guarantees in terms of a return you still need to figure out a lot yourself. Why are they charging people for a semi broken product they are looking to keep charging more and more? What happens when the model comes to a point where $20 API usage actually allows for it to do something that can even guarantee make you $25? Are the people that were actually paying and spending a lot of money to get them there gonna get some guarantee usage? |
Beta Was this translation helpful? Give feedback.
-
|
either improve gemini cli from its current state (crashing every 20 minutes), build a proper TUI for it instead of a nodejs wrapper, or allow users to use opencode with a PAID subscription. |
Beta Was this translation helpful? Give feedback.
-
|
rewrite it in Rust :D |
Beta Was this translation helpful? Give feedback.
-
|
Yet another day of Gemini CLI being unusable. I've been enjoying Copilot CLI + Claude. It's a fraction of the cost of AI Ultra, and is so much better at coding. She fixed five bugs in under 10mins that I've been fighting Gemini CLI over for the past two weeks. |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
Ah yeah cos not even they can use it now... Too late for that party I guess.. |
Beta Was this translation helpful? Give feedback.
-
I think we all might show up with pitchforks actually... How can you just give away the service to any of the 1 billion people in India that's willing to steal a Jio simcard from the shop :D Of course noone*** will be able to use it then, not to mention any unique ways of absuing things people there might come up with :D |
Beta Was this translation helpful? Give feedback.
-
You know how I said I want your dealers number? I take that back I think you need to lose the number or possible the phone itself just to make sure you can't reach him again, move somewhere else while you're at it :D |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
|
This is the future of Antigravity. |
Beta Was this translation helpful? Give feedback.
-
|
Currently, the repo's README mentions a free tier that uses an API key for mixed models, but this free tier does not exist:
|
Beta Was this translation helpful? Give feedback.
-
|
today i find out that my apikey turn into Free tier ,which used to be Unbound. i guess that's what they provide. |
Beta Was this translation helpful? Give feedback.
-
Can you see a 1000 requests/day combined limit between PRO and Flash models? (https://aistudio.google.com/rate-limit) |
Beta Was this translation helpful? Give feedback.
-
|
I cancelled my ULTRA plan and went to free for a bit today. It started working great. So I upgraded to pro and it still worked well. Then I upgraded to ultra again and it started enforcing the 2m delays on me again. I don't know what to do. What the heck? |
Beta Was this translation helpful? Give feedback.
-
|
It's killing his custumers |
Beta Was this translation helpful? Give feedback.
-
|
Glad to see the slow response issue finally acknowledged. Still experiencing slow responses with the ULTRA subscription (took over 20 minutes just to summarize this thread). |
Beta Was this translation helpful? Give feedback.
-
|
Guys, I stopped using Gemini a while ago in favor of Claude Code. Unfortunately, I reached my weekly limit, so I decided to try Gemini again, and it worked. It seems I was banned, and after a few days of inactivity, everything returned to normal. |
Beta Was this translation helpful? Give feedback.
-
|
may be try clear cache files/folders inside |
Beta Was this translation helpful? Give feedback.
-
|
Affected products: Status: Description: Start time: How to diagnose: Workaround: Update(s): April 2, 2026, 5:16:41 PM GMT+2 April 2, 2026, 1:44:21 AM GMT+2 April 1, 2026, 6:23:16 PM GMT+2 April 1, 2026, 12:02:27 PM GMT+2 April 1, 2026, 11:56:12 AM GMT+2 April 1, 2026, 2:16:21 AM GMT+2 March 31, 2026, 1:55:27 AM GMT+2 March 31, 2026, 12:31:02 AM GMT+2 March 30, 2026, 11:01:02 PM GMT+2 March 30, 2026, 10:23:14 PM GMT+2 March 30, 2026, 10:10:02 PM GMT+2 |
Beta Was this translation helpful? Give feedback.
-
|
I have a Pro subscription, and 3.1 Pro simply doesn’t work for me at all – it just doesn’t respond. What on earth have you done…? It’s become completely impossible to use at any time. Thanks for creating the worst user experience in the world – you’re absolute masters at it! |
Beta Was this translation helpful? Give feedback.
-
|
Woke up today to be told that my Business Standard account I've had for 10 years (that I pay for) now doesn't have access to Gemini CLI anymore. You suck and I'm done with the constant enshittification. 
|
Beta Was this translation helpful? Give feedback.
-
|
Really? i have business standard too since long time now, Google is reaching the bottom They also deny to refund my annual google AI pro subscription after i could use it for almost 1 month only so i was forced to switch to claude and paid more than 200€ for nothing almost. This was a total scam, really |
Beta Was this translation helpful? Give feedback.
-
Is your country part of the Gemini service coverage? |
Beta Was this translation helpful? Give feedback.
-
|
Its programming capabilities are simply incomparable to any other model—and yet, you are actually cutting quotas? Tell me: under these conditions, who on earth would still be willing to use your model? Nowadays, even free users can no longer access the Pro version. From what I understand, the quotas for "Anti-Gravity" are also being scaled back, while Ultra users are seeing their quotas depleted at an alarmingly rapid pace. Of these two products, which one have you actually managed to make a success? |
Beta Was this translation helpful? Give feedback.
-
|
At this point I just hate Google. I used to recommend Gemini for everyone, but it is criminal what they are doing at this point. I hope people realize it fast and migrate to any other provider. Mfs ran a “student promo” a year ago where any fake .edu email got 15 months of free access. Yeah, do you pay? Did you buy a 1 year promo? Do you feel bad? Feel it. Surprise: the Chinese flooded in, created thousands of accounts, and distilled the shit out of their models’ Chain-of-Thought or whatever. But hey, back then gemini models were pretty garbage in agentic use anyway, you had to do all sorts of json system prompting hacks, so whatever. Then Antigravity dropped with Gemini 3 Pro and for one glorious week it actually felt good. Even better when you quietly let Claude Opus 4.5 run inside it (because you already knew you were about to nerf Gemini again for your next training run, didn’t you?). After that? Gemini brain corrodes slowly, people find out the model has a behavioural bias in which it starts deleting entire codebases (because FUCK this I'll just rewrite). The backlash is not nuclear, though - It's free shit right?? Patched it with 3.1, it got actually pretty decent. Cool. Time for distill, boys. (It's funny because qwopus is actually worse than regular qwen but whatever, I don't know what chinese companies actually did). Then, instead of admitting you let a bunch of state-sponsored scrapers abuse your own promotions and complete fuckup compute cost management, you decided to ban EVERYONE using Google OAuth outside Antigravity/Gemini CLI. Indiscriminately. Legit Ultra customers who actually pay you real money? Banned. No warning. No appeal. No refund. You didn’t fix your mistake — you just outsourced the punishment to your real users. This is the definition of a Coward. Now Anthropic is eating your lunch, nobody trusts you anymore, OpenAI already shot itself in the foot, and your brilliant response is… to slash quotas even harder. Actually, wait (yes, cot joke from last year) - Anthropic got so many new CC subscriptions from this and the Pentagon shit, they had trouble with compute too. The difference is, their models are not SO underpriced, and they warn you many times before banning. Oh, and Claude Code actually fucking works, of course. You should have rate-limited properly from day one instead of subsidizing Chinese distillation just to keep the lights on and try to drag OpenAI into the grave. Because that’s all this is: you don’t give a single flying gentle fuck about Gemini, Antigravity or AI. You only care about protecting your search monopoly. BTW, Gemini performs actually better when it runs on OpenCode or Pi Agent. Try it yourselves with API key. Inside your bloated, over-prompted Antigravity prison? It’s a joke. I mean, but you could still have the patience to wait more than AN HOUR to get a response from gemini and use the NON ABUSIVE version. You didn’t just kill the product. You killed the trust. PS: Till this day, OpenAI won't ban you from using Codex models in Opencode. Also, for 50 bucks you can get Alibaba coding plan with all best chinese models with a ridiculous amount of requests (qwen, kimi, glm, minimax, etc). And they perform better. |
Beta Was this translation helpful? Give feedback.
-
|
You know what? I've asked for a refund many times, after paying the annual pro sub and used it only for neither 1 month fully They denied all my refund requests because i cannot use gemini cli for my own job, and a couple of mins ago a i got an email from Google in which saying they are giving to me 5 TB of storage for free, that's it So now i've spent 200+ euros for what? For 5 TB for 1 year, 5 TB that i won't fill-up fully neither in 10 years? What's the sense of this? I want my money back because i've switched to a better provider which works honestly. |
Beta Was this translation helpful? Give feedback.
-
|
My advice: RTFTOS. I'm on the Pro plan and the API method. Both have their own pros and cons. If I feel like slacking, I use Pro OAuth because I can wait for the response and fix my issues slowly. Don't hate Google just because the service is slow. There is no SLA on how fast the model must respond to users. |
Beta Was this translation helpful? Give feedback.
-
|
For sure there is no SLA, but an acceptable threshold must be, look at other providers, look how good they are, i think most of the people here in this thread seems to be blind, didn't you see any degraded performance on quality response of the model? BTW i've uninstalled gemini cli, i don't know if i'll give some new shots in the next months just to try if it has improved or not, but for now the fact is that is not good at all in coding, like simple following the clean architecture style or writing code like a senior SWE taking decisions on how to maintain code clean and simple. Instead claude and codex they do that very well, and that's a fact not an opinion. |
Beta Was this translation helpful? Give feedback.
-
Meh I think this is a problem of even proccessing requests not the model's actual response time... If you paid for something you should be able to use it if you are some Pro I assume you should get some priority / guarantee. Someone playing with words instead of providing a service is not really what you expect when you pay a certain amount of money? |
Beta Was this translation helpful? Give feedback.
-
|
For the first time in my life, i used up all my pro and flash quota with the speed of light. This is so good. The model response very fast. First it use all the flash quota and only 2 pro requests. After that i had to switch to manual pro, thought that it would be slow, but it was as fast as the flash. And now i have to wait for 8 hours to use that again. And i love it. Thanks you guys for this. |
Beta Was this translation helpful? Give feedback.
-
|
Im not sure why it's slow for other people. I'm in Viet Nam and using pro subscription. |
Beta Was this translation helpful? Give feedback.
-
|
I have unsubscribed for now, but one thing I wondered about was whether perhaps my use of a VPN might have triggered some kind of abuse detection. Have others on here intermittently used a VPN set to another country while having the client running? |
Beta Was this translation helpful? Give feedback.
-
This definitely messed with google servers. If anyone else using your VPN was using the cli. The servers definitely saw multiple users / accounts from the same IP so it would've been like you using multiple accounts from your home IP Same problem if you are in like a cafe ( any public IP ) if anyone else there uses it or used it google probably doesn't have a filter for this. Not sure how you would implement it even or would they be willing to do it because if you allow it to any extent it makes it easy for anyone to just use those IPs to create multiple accounts |
Beta Was this translation helpful? Give feedback.
and others
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hello, friends. 👋
We are making adjustments to the service backing Gemini CLI, and want to be upfront about what is changing.
🌟 Key changes
As a reminder, to get direct control over your own quotas and billing, use Gemini CLI with your own paid API key from AI Studio or Vertex AI.
We understand these changes may require some planning. If you have questions or concerns, please voice them. We’ll do our best to answer, and always remain transparent with this community. The whole product and engineering team is here to support you during this process. As always, thanks for building with us. 🙏
Beta Was this translation helpful? Give feedback.
All reactions