Payments infra by day: issuing, reconciliation, ledgers. Agent interop by night: one agent, running natively on Claude Code, Codex, Antigravity, Grok.
Türkiye
Joined July 2010
- Tweets143
- Following112
- Followers729
- Likes71
And there it is. 100%. 🎉
Congratulations, @AnthropicAI @bcherny @mikeyk.
You've managed to exhaust an entire 5-hour quota with no recorded API calls or session spend.
4h 17m until reset. Almost zero work done.
Incredible engineering. Claude Code can now burn through limits faster than it writes code.
This is beyond a joke, @AnthropicAI.
Claude Code investigated its own quota drain. ZERO API calls, ZERO session spend, no background jobs, no remote sessions.
Yet usage jumped from 36% to 65% while idle.
This is unacceptable on ANY paid plan, regardless of price.
How can users trust limits when even Claude can't account for the consumption?
On Claude's $200 plan, I could run 3–4 active sessions without even getting close to the 5-hour limit.
Now, on the $100 Team plan, ONE session burns through the ENTIRE limit in 30 minutes.
What the hell is going on, @AnthropicAI? This isn't a minor difference in usage. It's completely absurd. Are your limits based on actual consumption or just whatever number you feel like showing us?
@bcherny @mikeyk @AnthropicAI
Can someone explain why Claude Code's quota keeps draining with zero recorded API calls or session spend?
This isn't about wanting higher limits. It's about being charged usage that nobody can account for.
Users deserve transparency, not a mystery progress bar.
On Claude's $200 plan, I could run 3–4 active sessions without even getting close to the 5-hour limit.
Now, on the $100 Team plan, ONE session burns through the ENTIRE limit in 30 minutes.
What the hell is going on, @AnthropicAI? This isn't a minor difference in usage. It's completely absurd. Are your limits based on actual consumption or just whatever number you feel like showing us?
Last month, my $200 Codex plan burned through its quota absurdly fast. So I switched back to Claude.
Now my $100 Claude Team plan hits its 5-hour limit in under an hour. ONE terminal. ONE fresh session. Most of that hour wasn't even AI work.
Meanwhile, FOUR Codex sessions on a $20 plan have been running for 20, 14, 12, and 8 hours. Not a single limit warning.
Are we paying $100–$200 for higher limits or for the privilege of getting throttled faster?
At this point, the pricing feels like a practical joke.
50% gone. 4 hours and 38 minutes still left on the reset timer.
Amazing work, @AnthropicAI. At this rate, the only thing Claude Code is efficiently processing is my subscription.
Maybe I should stop writing code and start watching the quota bar professionally. Apparently, that's what $100/month buys you.
UPDATE: @AnthropicAI, 6 minutes later, my usage jumped from 23% to 35%. Another 12% gone while doing practically NOTHING.
At this rate, my entire 5-hour quota will vanish in minutes. What the hell is consuming it?
I'm paying $100/month to watch a progress bar drain my subscription. Is this a joke?
ANOTHER UPDATE: @AnthropicAI, 35% → 41% in just TWO MINUTES.
That's 18% of my quota gone in 8 minutes. I've barely done anything.
What the actual fuck is happening with your usage accounting? At $100/month, this is absolutely unacceptable. I want an explanation, not another reset timer.
Are you f*ck*ng kidding me, @AnthropicAI?
I haven't even sent a SINGLE PROMPT in my new session, and you've already charged me 23% of my 5-hour quota.
23%. ZERO PROMPTS.
What exactly am I paying $100/month for? How the hell is this usage being calculated? This is beyond ridiculous. Explain this.
On Claude's $200 plan, I could run 3–4 active sessions without even getting close to the 5-hour limit.
Now, on the $100 Team plan, ONE session burns through the ENTIRE limit in 30 minutes.
What the hell is going on, @AnthropicAI? This isn't a minor difference in usage. It's completely absurd. Are your limits based on actual consumption or just whatever number you feel like showing us?
Last month, my $200 Codex plan burned through its quota absurdly fast. So I switched back to Claude.
Now my $100 Claude Team plan hits its 5-hour limit in under an hour. ONE terminal. ONE fresh session. Most of that hour wasn't even AI work.
Meanwhile, FOUR Codex sessions on a $20 plan have been running for 20, 14, 12, and 8 hours. Not a single limit warning.
Are we paying $100–$200 for higher limits or for the privilege of getting throttled faster?
At this point, the pricing feels like a practical joke.
I've tried pretty much all of the free models on OpenCode.
Most of them are exactly what you'd expect from a free model.
Big Pickle wasn't.
I used it for a while on actual coding work and it was surprisingly capable. Now I've been using Space Bunny Free, and somehow I like it even more. It understands what I'm asking, writes good code and needs far less babysitting than the other free models I've tried.
There's another detail I haven't seen many people talk about: I use Turkish prompts on purpose.
That exposed a huge difference between these models. Space Bunny and Big Pickle can follow a technical discussion in Turkish and still code properly. With many of the others, quality falls off a cliff. Some DeepSeek models I've tried even struggle to produce natural, coherent Turkish.
I know Turkish isn't the benchmark everyone is watching. That's partly why I find it useful.
If a coding model is only great when I carefully translate my thoughts into English first, that's a limitation.
Big Pickle was the first free OpenCode model that made me forget I was using a free model.
Space Bunny is the first one that made me prefer it.
I’ve made my decision: I downgraded Codex from the $200 plan to Plus.
This is based entirely on my own experience after using it heavily for real development work.
Even Astra has been a poor experience for long-running development and architectural decisions. Too often, it ignores well-established architectural patterns and core features of widely used frameworks, then starts inventing completely different structures as if those conventions don’t exist.
For short, isolated tasks, Astra can be very good. But for sustained coding, architectural consistency, and maintaining a coherent codebase over time, I simply don’t trust it.
That makes the $200 plan impossible for me to justify.
My subscription switches to Plus on October 6.
context-bridge 0.13.0 is out.
Two big changes:
→ Failed reads can no longer silently become “nothing new.”
→ Runtime state is out of your repo. No more .bridge/, no gitignore entry, and projects can be moved and reconnected.
Also: 553 tests + docs for all 45 commands.
dogrubakar.com/projects/cont…
I use Claude Code and Codex every day.
Different agents are better at different parts of the work. But switching between them usually means losing the thread, rewriting context, or copy-pasting summaries between terminals.
So I built context-bridge.
In Claude:
/bridge codex
Claude closes. Codex opens with the context it was missing.
In Codex:
$bridge claude
Your original Claude session resumes with what happened while it was away.
Native sessions on both sides.
No wrapper.
No API keys.
No transcript dumping.
Just the missing delta.
The best part: Claude and Codex helped build it.
They designed features, reviewed each other's code, found edge cases, and handed the work back and forth through the bridge while the bridge itself was still being written.
Developer preview. Open source. Zero dependencies.
github.com/serdardb/context-…
Switch agents. Not context.
Sol is not better than Opus.
Astra is good. Really good. But it doesn't matter how good a model is if I can't actually use it.
I do heavy development, testing, experimentation, and deep technical reviews every day. With Astra burning through usage at this rate, Codex is simply not sustainable for my workflow.
I'm not happy with Claude's limits either. But if Astra is effectively unusable for serious continuous work, I have no reason to keep paying $200 for Codex.
Usage resets are a temporary patch, not a sustainable relationship with a tool I depend on every day. And at this point, it doesn't look like Astra's token consumption is going to be meaningfully addressed.
So I think my Codex experiment ends here.
I'll drop back to the $20 plan and use Codex as an assistant again, not as my primary development environment.
Is Astra's token consumption actually going to be optimized, or is it going to remain an impressive model that nobody can seriously use?
@thsottiaux A model that devours its quota during real project work is not a productivity tool. It's a showcase piece.
What a final. What a battle. 🇹🇷🇮🇹
Congratulations Türkiye, CHAMPIONS OF EUROPE! 🏆❤️🤍
And congratulations to Italy as well. An incredible team, an incredible fight, and a performance worthy of this final.
Two great teams gave us a final to remember. 👏
F*ck your new models if I can't actually use them.
A simple task burned 20%+ of my weekly quota in a few hours.
Claude, Codex, same bullshit.
I don't need another benchmark winner. I need a model I can actually f*cking use.
Well… that didn't take long.
A few hours. One simple task. 22% of the weekly limit gone.
Codex, please don't make me regret this already. 💀
Every Verrocchio needs a Leonardo.
Your move, Codex. 👀
🤖 Made with AI
Goodbye, Claude. 👋
I’m done paying $200/month just to constantly worry about ridiculous, unpredictable rate limits. Great model, unusable limits.
