The quota difference between the Gemini $99 Ultra subscription and the $199 version is absurd. I don't understand why Google is doing this, but the Pro subscription is essentially useless, and the Ultra version for $100 offers so much less quota. I really can't wrap my mind around that. They should at least tell the users it. Because, as far as I can see, they are not quantifying the actual quota differences specifically for Antigravity. I had two senior advisors, five implementers work on the codebase, and one QA agent using my entire platform for many hours to collect UX and UI issues. It started yesterday evening, after I upgraded to the $200 package, a decision that proved far superior. The $100 subscription would have exhausted its quota after just 1.5 days of the same work intensity, leaving me at only 16 % of the weekly quota. After upgrading, the agents worked all night in Argon on high, continued today, and ran an extensive QA session that simulated real users on my live production platform, creating accounts, receiving emails, and starting from scratch. I wrote a skill and a prompt for that QA agent, and I’m blown away by how well it works. It’s the first time I’ve had such an effective QA agent. The same prompt failed on the AutoClaw platform of GLM (which mimics Grok) and on AntiGravity with the Flash model (Flash 3.8 on high). In contrast, Argon completed a two‑hour perfect human‑QA simulation. I’m about to write a loop that integrates this process into my Constellation senior advisor/implementer, then runs a human QA tester. So far, only 8 % of the weekly quota has been used, and no cheaper model was involved. When I call sub‑agents, I force them to run Argon on high. In my view, the $200 Geminu Ultra plan is the best deal for coding. It outperforms Anthropic and OpenAI, offering the most reliable and capable coding model available. This level of autonomous testing matches what I’ve only seen from OpenAI’s Codex. It might also be due to the workflow harness that I'm using, the extensive agents.md file, and the local project memory that my agents.md instructions automatically create and update after every action. I saw yesterday someone publishing tips on working with Argon, and I saw there that it was specifically said that Argon loves instructions and also loves the task to test its work and to get detailed test descriptions. This would make it essentially the absolute opposite of the showboat models of OpenAI and Anthropic. My workflow harness combined with Argon might be a match in heaven. I can't quantify how many magnitudes better than anything that I've experienced since I started using AI for coding. I’m considering packaging all of this and putting it online, if I find the time.

I'm really wondering why I see so many people complain about the Gemini 4 Argon quota. Testing it, obviously, it seems that, in a Pro account, the discrepancy in quota between Pro and the two Ultra account subscriptions seems to be huge. I'm a long-lasting Gemini user. I was, until a few months ago, on the $200 plan because I was using Opus 4.6 back then until the quota was extremely limited, and Gemini 2.5 was not a very good model for coding. That's why I reduced my plan, but I'm using many other Google applications that are in the Gemini bundle, so I stayed with them on the Ultra subscription for $99. The Ultra plan is a pretty nice package, specifically if you like to have a bit more cloud storage, but also YouTube Pro and things like that. Now, after testing Argon for3 days, I worked intensively with it. It's the third day in full, with several sessions, and my weekly quota was down to 16%. Now, a few hours ago, I upgraded to the $200 subscription, and looking at how slowly the weekly quota and the 5-hour quota are draining, I would say this is definitely more quota on the $200 plan than you would get for an Anthropic Max plan. I'm absolutely positive that this is a much better deal because it's also a much more reliable model, specifically when you have projects where you don't want to break something while you implement something new (like it will happen with the other so-called frontier models). I didn't have time until now to go into creative discussions with it to see how strong it is as a thinking partner, but as a coding agent and senior software engineer, it's really stunning.

It was clear that this would start sooner or later. They will ban all the Chinese models to protect the business model of Anthropic and OpenAI.

Kudos to the @GoogleDeepMind team. Gemini 4 Argon is a hell of a model. It's incredible. For quite a while, it's been the first model where I have the feeling that, even with strictly regulated QA and auto testing involving professional implementations harnessed by strict workflow instructions via agents.md, it's a model that I can blindly rely on. It came just in the right time because, I was in the finishing phase, polishing and doing large-scale quality and security testing of a project that I worked on for more than a year. After the experience of the past two days, I will definitely upgrade my second-class ultra account again to the first-class ultra account because I will need the weekly quota. That's the only model that I'm willing to fully trust right now! It's really a huge relief for myself because, even with Anthropic models and GLM, I literally read almost every reasoning line of the output, because sometimes I read really freakish things. I often had to interrupt the models in actions that would have been, not always catastrophic, but definitely regressive. After working 3 days and getting stuff done that would have needed 10 days in GLM 5.3, or a week with Anthropic models, I can really relax, sit back, and watch Gemini 4 Argon do the stuff really quickly and reliably. It's an absolute blast to work with this model. I would not trust the Gemini Flash model for this type of work. Definitely not, but Argon is in another class. Google really delivered with it. I don't care what some one-shot wipe coding 3D gimmicks might look like in Argon compared to Opus 5.5. I worked with all of these models, and I know whch one to pick for relaible coding. Nothing comes close to Argon for serious engineering work. It's not even close. I am not comparing model intelligence. I'm comparing reliability. To me, the ability and also willingness of the model to tightly follow a proven workflow framework is the most important aspect of reliability. Exactly this is where the Anthropic models started to fail big time. Anthropic tried to whitewash this by saying, "Hey, you need to adapt your agents.md files and use different prompts," but I call BS and think there is a deeper structural flaw in their RL process (and also in them polishing their models for more spectacular vibe coding and one-shotting of things). I'm so glad that Google did not do this. GLM 5.3 is better at following instructions than the Anthropic models, but due to the shallower depth and intelligence that it has, you cannot compare it to Argon. Right now, I also have to revoke my recommendation from the GLM Coding Max plan, definitely in favor of the Ultra plan of Google, because if you really want structured, fast, reliable, professional code engineering, Gemini 4 Argon is the way. BUT.... None of the things that I did in the past 3 days, I could have done so successfully with a lobotomizing context cap that Antigravity has on the Google models: 230 thousand tokens. Did the Antigravity team forget this limitation from previous models? Or are you out of your minds and this is intentional? I patched it to 980,000 tokens, and Argon is by far the best senior engineer that I have worked with, let's say, for 2 years or so, but only without the context cap. If you are not aware of it, fix it. I mean, not for me. I solved the issue, and if any Antigravity user wants, I can have Gemini write the recipe for how you can patch it.

Google, respectively, the Antigravity team, thinks it is a great idea to not even inform you about the fact that they are lobotomizing and capping their models in Antigravity. You get only an effective context window of 200,000 to 230,000 tokens although their models have more than a 1 million-token context window. It's absolutely hilarious. I could not believe it, but I definitely verified it, and I also hacked it meanwhile.

After several months of testing Anthropic's Claude until they started fucking it up with Opus 5 and Fable 5.1, I went back to GLM 5.3 in Z code. Today, I thought it was the right time to test Gemini 4 Argon in Antigravity because I'm finalizing a big project that I was working on for 12 months. It turns out the model is really capable. I didn't expect that. It really loves detailed work. It has essentially the intellectual breadth of the Anthropic and OpenAI models, so it's superior to GLM 5.3, clearly. This is a huge difference: it did not lose the ability to actually follow a tight workflow harness. For someone like me who has developed a workflow harness that helped him develop larger projects over up to 16 months, with massively growing codebases, the only thing that kept my sanity and kept the projects afloat was essentially a tight workflow framework and specific standards, procedure, and protocol. Beginning with Opus 5 and also Fable 5.1, Anthropic and OpenAI models totally lost the ability or the willingness to actually follow workflow instructions, so they got very sloppy. They became great vibe coders and one-shotters, but they became horrible senior engineers who work reliably in a bigger, established codebase that needs targeted changes or changes based on absolute quality control. No implementation without sufficient QA. Now it turns out that Gemini 4 Argon seems to be the best of both worlds. It's definitely more capable than GLM 5.3. It has a huge depth. It has the intelligence, and what made me really happy is that it actually really follows workflow instructions and process. Right now, to me, this is the ideal model for what I'm doing, and I really couldn't believe it when I saw this. Antigravity, are you actually really serious in lobotomizing your best model in your own IDE to a 230,000-token context window? I mean, are you fucking kidding me? Do you live in 2024? Is this intentional? Is this due to limited compute, or did you just forget to upgrade this in your IDE? You run huge promotions that your model has a maximum output of 1 million tokens. What the fuck? Who the fuck needs 1 million-token output? You are advertising a 1 million context window, and then you are fucking lobotomizing your model at a quarter of the context window. Are you fucking serious? Fortunately, I could immediately hack it and patch it, and now I have the full million context window when I work with Gemini 4 Argon. Why the fuck would you do that? Are you stupid? Yeah, I almost forgot you had another clamp on it. When it comes to loading agents.md files into the system prompt, this was also capped to 24 KB. I mean, are you fucking kidding me? Anyone who works with Antigravity and wants the patch, just comment. I will publish it here. Won't work with Opus 5.5, though, because this goes over Vertex. The patch only works with the Google models. Opus 5.5 is even more limited than Gemini 4 Argon .

Just for the record, this is also France, when young farmers dared to protest.

As if France weren't already on the brink of financial collapse, the cancer of Islamomsrxism has now also raised its ugly head. Europe is done. Why does the leadership show this extreme devout cuck attitude? Do they want France to implode?

This will be a fun time to be alive, when all the people who built stacks for 60k to run some quantized, shitty, lobotomized model finally realize that their stacks are worth one-tenth of what they paid. I stopped my ambitions to purchase a capable stack for at home when I saw that China started to solve the chip issue for itself. Because China cannot do small volumes.

Imagine being a Google Ultra user and getting earlier access to Claude Opus 5.5 inside Antigravity, but not getting access to Google's hugely announced (I think it's already two weeks ago that they announced it) Gemini 4 Pro. A model that I probably would not use anyway, but this is so hilarious: they announce it, give it to a few shitfluencers to hype it up, and then even as a Google Ultra user, you don't get access to it. Google is definitely, it seems, too big to fail. It allows itself these hilariously bad product teams that act with comical strategic frameworks and still doesn´t go broke. Sometimes you think, are they doing this intentionally to harm the brand or their own products?

Berlin steht im wahrsten Sinne des Wortes eine Zeit des Aufbruchs bevor. Was eine sozialistisch-kommunistisch-marxistische Regierung in 15 Jahren alles schaffen kann.

It would be interesting to know: 1. Which models and IDEs you use 2. How many tokens you turn around every week I am using ZCode. The screenshot shows the models.

Does this even constitute the level of a letter of intent, or is it just the usual dog and pony show? It certainly doesn't look like a binding contract with actual responsibilities. Also... if you look at the fine print, it looks more like a dog and pony show: the intern who wrote this wrote under the signature of the President "Unites States".

They can't even hide it that they are ripping you off when you use their API. OpenAI's Tibo right now more or less admitted that they might be inflating API prices to make the subscription look much cheaper. Why would you even think of something like this, or voice it in this form, saying that you are not doing it? Why would you even indicate that subscription prices are somehow made in relation to API prices? Saying, we might have underestimated the amount of subscription users that they have compared to the amount of API users that they have, is a very strange statement to make. Imagine that: it doesn't make any sense. Why would you say something like this? It means you had this thought in the past, you have it now, and you would do it. Even if we give them the benefit of the doubt and assume they did not do this in the past, they are using it now as an argument for why they need to halve the quota they are giving you, because they don't want to inflate the API prices. Do they not have any API customers? Would you not be quite pissed if you were using their shitty models via API, if you were stupid enough to do that? I mean, how crazy is this? Why would someone say something like this? OpenAI's Tibo right now: "(c) We don't want to put an incentive on ourselves to artificially inflate the API list prices to make it look like you are getting a lot (and workaround it through discounts, etc)."

Der Bürger soll sich mental, wirtschaftlich und physisch dem Staat unterordnen. Krieg löst jedes Problem des Staates auf einen Schlag: - Er liefert den ultimativen Feind, der jede Kritik als Verrat erscheinen lässt. - Er rechtfertigt Steuererhöhungen, Schuldenorgien und Geldmengenausweitung, die im Frieden sofort auffliegen würden. - Er erlaubt Wehrpflicht – also die Zwangsrekrutierung von Körpern und Lebenszeit. - Er zentralisiert Macht, weil „jetzt keine Zeit für Diskussionen“ sei. - Er schafft eine künstliche Einheit, in der der Einzelne zur Ressource wird. Im Krieg entsteht eine Hierarchie der Werte mit dem Staat an der Spitze. Kunst, Wohlstand, individuelles Leben, Vernunft – alles wird nachrangig. Für die geopolitischen Fantasien von Leuten, die selbst nie an die Front gehen, sollen wir sterben. Der Leviathan braucht die Kriegsstimmung, weil sie ihm das gibt, was ihm im Frieden fehlt: unbedingte Loyalität, Ressourcen ohne Gegenwehr und die moralische Lizenz, den Einzelnen zu verheizen.

You can freeze and slowdown as much as you want. I’m using Chinese models anyway. And China won’t slow down. The US ran out of compute; that’s why the labs are lobotomizing their models and all the plebs get is a lame influencer propaganda shitshow. They’re keeping the good stuff for themselves to connect wet labs and invent the future while morons do their doomer bidding for the AI oligarchs. This century will be China’s century. They’re just starting to build compute, and their models were already at US levels with 1/100th of the compute. Nobody needs the nerfed shitshow coming out of US labs. Only idiots still send them money. Have fun with the slowdown. It’s already happening, but most of you are too stupid to tell the difference in the models’ IQ. They are not giving you the intelligence that they could give you. They never will.

Chinese models are the future. Quit US lab subscriptions. Quota pressure will only get worse and the models will be compute-starved. US labs are a dead end. Where US labs must ration quota and intelligence, Chinese models need far less compute and are meanwhile getting more of it. China has built its own chips and is building HUGE datacenters. Their performance will only increase while the lousy so-called frontier models from US labs get worse. Stop financing the mediocre engineering and idiotic business models of these hot-air corporations. Pierce the bubble. Free yourself. Go with the winners. China will dominate the next cycles.

Oh, they will slow down AI. But not the one they use for world domination. They will keep accelerating hidden models that create new science and help them win the future. What the AI oligarchs want to slow down is the AI they give us. People like Tao should not be naive and should be more explicit in their demands. Anyone trusting Altman and Amodei is out of their mind.

You literally can't hate Misanthropic enough... they pretend that more intelligence of the models is a safety issue while in the background they are creating the real potential hazards. What Anthropic actually wants to do is to slow down your options, slow down the models that you will have access to, while they will increase the intelligence of the models they use as much as they like. And not only that, they will then have their AI work autonomously in wet labs. What could go wrong? They want to own all the future technologies, all the future materials, all the future businesses and markets. Stop sending them your money. Every dollar that you pay them supports their nefarious plans. It's an evil company. The more a corporation pretends to be philanthropic, the more the actual opposite is true. Stop supporting the self declared altruists who think their call is to create a deity. The models that they are willing to share with you ain't worth it anyway. They make you believe for the first one or two days after a new release that the models are great. Then, because they are such bad business people and because they don't have enough compute, they usually have to ratio, restrict, and reduce compute allocations to their models. You're getting a shitty, unreliable quality, you're paying a crapload of money for it, and you are supporting an evil system. Enough of that.