Repository navigation
Service update: mitigating abuse and prioritizing traffic #22970
Replies: 401 comments 522 replies
|
these changes may require some planning = Switch to Codex btw 🤷🏻♂️ |
Agreed. I made most things just using the free tier... I'm glad I mostly finished the project that was, I think, important. I figured out Self-Driving before Elon Musk :D |
|
Google about to shoot themselves in the foot again. Guess it's time to look for other solutions. |
话糙理不糙 |
|
Here is a clever workaround for Plus/Free accounts getting discriminated: https://haryn.hashnode.dev/the-gemini-cli-paradox-route-to-your-own-endpoints-a-digital-thriller |
|
Thanks for moving to GO, node.jshit is ass, it uses almost 40GB when you have a serious workload running on your GPU with a local model. Then there's no need to use your enterprise trial and send request straight from pwsh and hopefully the experience will be even better not just performance wise thanks to GO, if you make the connection to AG IDE proper I see crazy potential in that! |
|
Is opencode via https://github.com/jenslys/opencode-gemini-auth usage safe? (Pro account) |
|
@mitchell-wallace if the argument applies to sama, why doesn't it apply to google? google has no interest in developers reputation/karma? |
|
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. |
|
@YurySolovyov google is big enough not to care. look at every single release they've made for the past two years. practically every single one (with a few notable exceptions) was half-broken, fragile, unsupportable headline-grabbers. it seems google has a bad case of split-brain, one side (the research they do, e.g. attention is all you need, turboquant, imagen, the inference TPUs, etc.) is slower but more impactful overall, while another side is moving fast and breaking things and getting little progress made on the way, (think the dumpster-fire that is Antigravity, the shallow feature-set of NotebookLM, the Gemini chat interface breaking down after 50k conversation length, etc.). At the end of the day, Google is a trillion-dollar company, and they aren't burning other people's investments like Sama is, they are already a huge company well before. They have the ecosystem, the integrations, the existing partnerships, and their own existing user-base, all of which they can choose to gradually burn if necessary to beat openAI/anthropic. not by necessarily being better, but just being "fatter". |
|
Let's say it doesn't take much effort to be able to become a web dev. compared to one of their true geniuses that created Gemma which is fucking sick, especially if you have a 4090 or 5090 or Ada card with 24GB+ VRAM. It's actually almost a genius even on a 4080. I could create web services with java scripts in Notepad as a kid. Could I create a super sick AI model like Gemma? Hell no! Also since AI came along, no wonder it was reported that 75% of Google's code was made by AI. Their only code that can't be replaced by their Gemma model is basically just their AI models and the backend infrastructure. Rest of their services are web code, Android is just a Java VM with a Linux bootloader that Gemma can code as good as it possibly gets. They have the infrastructure to run a model of Gemma that is smarter than anyone of their employees except for a few key people like the AI model team and infrastructure team that handle the backend and network loads that are so insane they had to make a special bypass of the Linux kernel because it's too slow to handle the communication between the server NIC and CPU. Google just needs to keep maybe 10k out of their 80k devs. All they need is ideas for future services and products and Gemma will handle almost all of the rest. Web dev in a year maybe already, but certainly no later than 2030 and onward will "just" be about designing UI's that are easy to use and look good, zero lines of code will be done by humans. And how much has the UI changed in Gmail, Google Docs and so on over 2 decades? Basically nothing so it's not like they need a massive amount of UX designer's either. Both Claude Opus 4.7 and Gemini 3.1 Pro can write perfect html, css, java, you name it. Especially Claude is actually insane, doesn't just almost always nail stuff first try, it creates proper documentation of everything and nice read.me and get started guides and tests it's own code on autopilot, you only need to let it free and it will let you watch youtube or what ever instead. |
|
u are right |
|
L move.
|
|
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. |
|
Google Plus subscription here in the EU: 
Its great that Google is capable of providing good and free open-source models such as Gemma 4; it sucks that they are behind on the cloud frontier models .. and treat Plus as second rank citizens 😞 |
|
|
It is not. I tried it weeks ago. It only gives you tokens for the web and nanobanana or something like that. I had to go back to pro. |
|
most of us here don't give a damn about the ecosystem, we just want the CLI. |
|
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 |
|
The readme doesn't mention anything about Flash quota. |
|
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 |
|
Its 150 requests per day with the flash models |
|
Good question. Our published limits include totals across model calls including Pro (when applicable), Flash and Flash lite. |
|
IDK, each request spends to much time queued that I never reach the quota with flash. |
|
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? |
|
Good question. Gemini Code Assist for Enterprise remains an option. |
|
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。 |
|
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. |
|
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. |
|
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. |
|
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. |
|
@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. |
|
+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. |
This is very vague can you provide more details which licenses get the higher priority and what defines account standing? |
|
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. |
|
seems like they aren't good in systemic thinking. weird. |
I thought they had those big thinking models. |
|
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. |
|
I work as a business consultant in addition to developing software, and i may also stop if these changes impact workflow whatsoever. The priority and model changes are especially concerning. This was google's chance to gain some clout by taking the high road and make millions of online businesses that connect to their existing paid services. If they botch this that will never happen again. |
|
Why not just explicitly block / ban users with openclaw? After reading the doc that seems to be the only ISSUE. |
|
Yeah, I think the problem is detecting when it's used by OpenClaw. |
|
Why?? If I pay for credits, allow me to use them as I want |
|
Ofc credits should be allowed. The ones with problem is OAuth, as I think so. |
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 |
|
a really bold and wrong move, i guess i'll start looking for alternatives |
|
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, |
|
Agreed for moving to qwen. Qwen as light model, Claude ad heavy ones. |
|
@AhmadMWaddah KiloCode has always free tier LLM, and it can be wired to QwenCode, which is one of the best alternatives at the moment. |
|
lol, lmao even, yep that's over. just ban openclaw users.
|
|
Why?? If I pay for credits, allow me to use them as I want |
Yeah it doesn't really make sense isn't that what the quotas are for lol |
|
LOL. |
|
Now the CLI is useless with flash models for free use. It's a shame... although blocking the lobster lovers would have been enough.🤌🏻 |
|
I agree. Flash models should not be anywhere near a terminal. |
|
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 :-(. |
|
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. |
|
China AI opensource disruption of US AI bubble seems to be working ;-) |
What the hell is wrong with you Google? |
|
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? |
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? |
|
After reading some of the comments I'm unsure, whether using a premium plan will be good idea. |
|
Use the pay-as-you-go API if you are unsure whether your usage will breach the TOS. |
|
Well, traffic prioritisation is working absolutely great! With almost no used quota and premium account, no models are accessible today. Or better said, every tenth attempt might yield an answer after waiting for 10 or more minutes. At this point I can do stuff faster manually. |
I have no problem to use Oauth to fix my brownfield project. I have API billing method for backup. |
|
|
This is wonderful progress. Google has successfully halted all abusive and legitimate inference traffic from using Gemini CLI during the time periods between Tuesday morning to Friday Afternoon. The answer was right there all along, just shed everyone's connections from the server clusters and deprioritize free, pro, ultra, tier 1 API, and tier 2 API users. Looking back, one must wonder why this wasn't implemented sooner! It is just so easy to ignore the valid requests to achieve all traffic prioritization goals! Good job Google!!! |
|
Absolutely fascinating to watch Google (finally!) having to deal with scale ! Scale of viral global adoption of their initial AI developer offering (which used to be leading only a couple of months or so back). (Anthopic's 40b ARR and growing - this is no longer our grandpa's "search scale"!) Hope this continued struggle will choke out some short sighted mid-level management and get it to attention of some higher up visionaries 🤞 |
|
can someone tell me if invoking "code": 429, on a Pro plan. There's nothing else in the error message that points to a possible restriction but the OP here is super vague so I'm not sure if they'd make it obvious in the error message if I were being restricted... |
|
Is the request itself that trigger the filtering. Gemini backend can see your prompts but won't use it for training. So if you use it as invocation tool by 3rd party then you will have risk to get ban. Gemini cli is tools for development and not for 3rd party automation. If you use for automations, please use API method. You won't get ban for using it. Oauth is value added service from pro or ultra plan with large cloud storage. |
|
This is such a letdown. I'm a subscriber to the Google One AI Pro plan, yet after sending just a simple |
|
how about via Antigravity? |
It seems ok in Antigravity. I tried again in gemini-cli with the |
I rather using gemini cli with api key than using something that is forked from it. If i need a feature I can develop it myself. Whether open source it or not is my personal choice. Please stop promoting 3rd party app in 1st party repo. Also, using vault will not make your software secure. |
|
100% agree, if you just want google only then use gemini cli. If you want to work with any ai and have your keys stored securely and swapped out as your model swaps out and you want the open source gemini project feeling then use bare-ai-cli or maybe a fork of that cli if it happens... Easy. Lots of other discussions on this forum about other ai providers too btw, Luckily it's a free world. If you don't like it then just scroll past , but I do want to thank you for your time. Honestly, I can tell who is over 40 and who is not by the comments. It's a free country/world. True Open Source projects used to mean something. For any other vibe-coders listening, just remember, that we all stand on the shoulders of giants. (not mag7) literal human beings who built most of the underlying software and I want to pay special thanks to the 600 contributors who helped make my project a reality (by abstraction) along with the thousands of others who those 600 built on. Thankyou, for reading. |
|
@Cian-CloudIntCorp dude it's still an ad, we fight google here, not look for alternatives. there is plenty of them, and nobody is looking for them here. |
@kstepyra, I do hear you "dude" :), but I did not have googles budget and the project is free, I WAS honestly just looking for some folks to try it and provide feedback and thought hey look "a bunch of disgruntled vibe coders mad at google would be a good start", but hey, live and learn. lol. All your points are taken, most folks here want "click and play" and for the "ai to brush their teeth too". That's Ok, I get it , however, the best way to fight google if that is your cause is to be model agnostic and to use ALL the AI providers. This is exactly what happened with anthropic yesterday, the deal with X was effectively a response to the community backlash over the low limits applied to their plans. Thank you for your feedback. |
|
Use any LLM with gemini-cli follow these instructions: https://haryn.hashnode.dev/the-gemini-cli-paradox-route-to-your-own-endpoints-a-digital-thriller |
|
amazing fix now gemini cli cannot authenticate - sms delivery just fails in Spain. im wondering what a point in such a great security improvements. no way authenticated besides sms is avaiable wtf ? |
|
killed by google |
|
Gemini CLI - RIP |
|
@doroved did you already tried it? It suffers the throttling request mechanism as well or it is now reliable and fast as other LLMs providers? |
I've already tried it, but the experience with Flash 3.5 or CLI is still unsatisfactory; after 2 hours, the server experienced high load. |
ok then 😄 i was curious to give it a second chance but you've confirmed nothing has changed since i've switched to claude 2 months ago now Thanks anyway! |
I just tried again. |
|
RIP |
|
https://developers.googleblog.com/an-important-update-transitioning-gemini-cli-to-antigravity-cli/
Gemini CLI is dead. Google killed it and replaced it with Antigravity CLI, written in Go. For everyone wondering why they didnt care.... thats why. They knew it was a dead end project months ago but couldn't say that.
- Thanks,
Steve Thompson
Sent from [Proton Mail](https://proton.me/mail/home) for Android.
…-------- Original Message --------
On Friday, 06/19/26 at 12:02 autom8ed.it ***@***.***> wrote:
> Hello, friends. 👋
>
> We are making adjustments to the service backing Gemini CLI, and want to be upfront about what is changing.
> 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. 🙏
I tried to help you fix your authentication system that is unable to identify sub-tiers, that issue is still open and not fixed even though I have given you step by step guide my mother can follow and fix.
Did you fix it, did you even give me a thanks? No you banned my account, every card I've used with your dog company.
Then I look for the fun of it if you've fixed basically anything, ofc not.
So you picked node.js as the base, make a cli that's completely broken, scam every consumer with a sub for months.
I dpm't know what more to say than you completely lost the plot if you think anyone still have any trust in you or your products.
MS Intelligent.Terminal which is just a fork of WT and still sort of "hobby project" get more updates/fixes per hour than you deliver per week. Also it's made in C so it uses 0 resources in comparison and it's free.
Given that you don't even fix issues when I make a step by step guide for you and instead ban my consumer account, all my cards and my API key so I had to use my brothers account to ge ta new API key (your security is a joke, just like your code) and continue where I where in 5min.
But I digress, the point being, give me one reason someone would spend a single second on this project when your work speed is at snail tempo, your "thanks" for helping is complete ban, the cli being a pile of turd.
Like what world do you guys live in? 2000micrograms of LSD I guess?
—
Reply to this email directly, [view it on GitHub](#22970?email_source=notifications&email_token=AJEV75RB24VMIXELEUHQIAT5AVPZNA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZTGY3DOOBTUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-17366783), or [unsubscribe](https://github.com/notifications/unsubscribe-auth/AJEV75V7CXAYZGL2IUX32NT5AVPZNAVCNFSNUABHKJSXA33TNF2G64TZHM4TMOBRHE3TEMJWHNCGS43DOVZXG2LPNY5TSNRYGU3TMM5BOYBA).
Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for [iOS](https://github.com/notifications/mobile/ios/AJEV75XPED4WGOSKX7XBWMD5AVPZNA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZTGY3DOOBTUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSVGM33PORSXEX3JN5ZQ) and [Android](https://github.com/notifications/mobile/android/AJEV75SVN2KVGIR3TUDGJML5AVPZNA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZTGY3DOOBTUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSXGM33PORSXEX3BNZSHE33JMQ). Download it today!
You are receiving this because you commented.Message ID: ***@***.***>
|
Sort of true, they ended support for consumers but they still update/support it for Enterprise. This can be verified by looking at the repo that has several commits last 24 hours, one just an hour ago:
Also they released a new version just 2 days ago: |
|
I'm still using it with API key and business plan token flexing using both tools. Gemini cli for pipeline, agy cli multi agent capabilities. |
|
The recommendation to use your own paid API key for direct control is the right one for power users. The thing it doesn't solve is budget predictability: quotas, capacity limits, and per-token overages still show up mid-sprint. For teams running agentic workloads, we found it useful to move the heavy, long-running work to a flat monthly environment where the bill is the same whether you use 10M or 100M tokens. We run UltraWork on that model (https://vibecodingagency.com/gpu-cloud/). It's not a drop-in replacement for Gemini CLI, but a useful complement when capacity limits and metered billing become the blocker. Disclosure: I help run Vibe Coding Agency. |
|
booooo |
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. 🙏
All reactions