User switches from ChatGPT to Claude after Astra quota runs out
Paying $100 a month, the author says Astra is capable but its quota runs out within days and Codex is nearly unusable, relying on Tibo's resets before finally moving to Claude.
Okay, here’s why I switched my membership from ChatGPT to Claude. Maybe OpenAI can take this as a lesson, or maybe they’ll simply ignore it. At the end I’ll also mention the good things OpenAI does that I think they should keep doing, but first, the bad things. The things that made me switch to Claude: 1. Tibo’s (@thsottiaux) resets. Yes, you read that right. I’ve been using ChatGPT since it launched, and my latest “membership spree” at $100/month started in April. Overall, it was a good experience. I had come back from Claude because its limits were just unbearable, while ChatGPT seemed quite generous with theirs... Until... GPT-6 Astra. A model so good that nobody had seen anything like it before. Extremely intelligent, really good at understanding what I’m trying to say even when I explain it badly, coding at a level never seen before. Its test results shocked me; they were on another level compared to any other model. However, this came at a cost. The cost was that the limits were running low all the time, but hey... It’s fine, right? Tibo keeps resetting them over and over again! How cool! But then, bam, he doesn’t reset them. Well, alright. I guess I’ll try some limit management. Even while managing my limits, I still run out of usage in just a couple of days... Limit reached. I’m paying $100/month (and yes, I understand that we’re still underpaying, but $100 is not the kind of money I can just throw away on something I can barely use) and I can’t use anything in Codex... A week passes, and I tell myself I’ll be even stricter with myself this time. A couple of days pass, and I still have some usage left... Then vague posts on X make it look like Tibo is going to reset the limits again in the morning for sure! We got used to it - vaguepost = reset! Let’s go. So I burn through what’s left. The reset never comes. Another stretch without any usage. Fuck. Okay. 2. Releases that shouldn’t have been released. Alright, GPT-6 Sol gets announced, and this one is really going to change things: much better limits, a model close to Astra, yet way more usage-efficient! Bam, it’s shit. It’s even worse than GPT-5.6 Sol. Why the fuck would they release a model that’s worse than its predecessor? On the very same day, Anthropic just casually launches Opus 5.5 (we’ll get back to this later, btw). An Astra-level model. Sure, you can argue whether Astra or Opus 5.5 is better, but I decide to try it on the $20 plan, simply because people keep talking about how good its limits are. I try it and damn. For $20, I basically get the same usage from a model of the same level as I was getting from Astra for $100. Alright, now it’s starting to look like I might leave OpenAI. 3. This is where we get back to the “Anthropic just casually launches Opus 5.5” part. Anthropic always do those casual launches, without screaming to your face, while OpenAI sometimes can feel like one non-stop hype train. So... OpenAI starts hyping up DevDay like crazy once again. It happens, and GPT-6.1 Sol is introduced. Yup, this one is quite a decent model with much better usage. However... Bro... The day before DevDay, Anthropic also released Sonnet 5.5. So now I have Opus 5.5, an Astra-level model with quite decent usage, while Astra is still barely usable. And I have Sonnet 5.5, which is even better on usage and still a very capable model... So what do I get from you, OpenAI? You know... for being quite loyal for quite some time. Ahhh, yes, some more resets, and Dots that I still can’t use to this day because I’m in Europe! Bam! Yeah, I get that this is a two-sided problem: it’s not only OpenAI’s fault, regulations play a role too. But why not wait a bit longer and release it to all of your members at once? Okay, I think maybe it’ll just take a couple of days and I’ll be able to try it before my membership expires. Maybe I’ll like it so much that I’ll actually renew my ChatGPT Pro $100 membership instead. But nope, time passes, and passes, and passes, and my membership expires... To be honest, I never got to test Dots at all. But I did test Grok Bot and found it barely useful for what I need, and I suspect I’d end up feeling exactly the same way about Dots. So this isn’t only negative: I find OpenAI’s team on X really cool. @victornunez always seemed genuinely invested in the issues I shared with him, @edbayes noticed my observations about the UI, and even @thsottiaux himself replied to my observation that Ultrafast wasn’t actually ultrafast in the API Playground, and that issue got fixed really quickly. I also still like Codex more than Claude Code overall. But that’s just not enough these days. Personally, I’d prefer OpenAI not to launch as many different services as possible. Sure, it’s nice to have image generation, but ChatGPT, image generation, Dots, Codex, GPT-Sites... sheesh. And GPT-Sites is the one I hate with all my heart. I probably interrupted Codex more often because of it than because of anything else: I kept having to tell the model to stop building things on GPT-Sites, with that freaking login every single time... I mean, yeah, Claude has artifacts, which I don’t like either, but at least I don’t have to log in every time I open one :D I’m also pretty sure OpenAI was even thinking about launching their own Suno-like music service at some point. Don’t waste the compute on all of that, put it into the limits instead. Solve AGI first, then monopolize everything :D What I want is simply the most capable model with the best possible usage. Anthropic doesn’t have image generation, a “bot”, music services or whatever, and I’m fine with that, because I don’t need any of it. Also, I’ve said this before and I’ll say it again... Why doesn’t any AI lab offer loyalty-based usage? Let’s say: stay for 3 months, get 15% more usage. Stay for 6 months, get 30% more. Stay for 12 months in a row, get 50% more. Something like that. In my eyes, it would be a win-win. Customers would become more loyal, you’d be able to forecast your revenue more accurately, people would be happier with their limits, and so on. And here’s one thing no LLM will tell you... What if your AI lab never gets to the point where AGI it pays for itself? :D Then loyal, long-term paying customers are exactly what you’ll need.