@thsottiaux And why doesn’t the support team respond properly to inquiries from clients who pay $200 for the opportunity to work 6 hours a week without seeing any significant results?
@melvindvivas I have no intention whatsoever of going back to OpenAI’s products; they completely let me down in September. Astra is a useless piece of rubbish.
@annapanart ChatGPT is absolute rubbish. In September, I used up two weekly limits and two restarts, but I haven’t made any progress at all. Now I’m only working in Anthropic Cloud – OpenAI can go to hell.
This isn’t a controlled benchmark. It’s more important to me than one: it’s my actual product. I spent weeks, huge amounts of Codex quota, a $200/mo Pro subscription and additional credits trying to move it forward. Then I changed the main operator. Two days later, the project moved. That’s a hell of a product comparison.
I finally changed the main engineering operator on Ashitaka.
Same real project. Same repository. Same overall goal and high-level orchestration.
Codex/Astra → Claude.
The difference after ~2 days is brutal.
OpenAI: 1% of my weekly limit left. Claude Max 20x: ~14% used.
Meanwhile, my OpenAI support case about the abnormal Codex usage is still unresolved after more than a week. I supplied timestamps, screenshots, reset reproductions and Analytics. Getting a concrete backend explanation has been painfully difficult. Public posts on X produced essentially no substantive response either. @OpenAI
@AnthropicAI@bcherny — credit where it’s due. Claude is currently giving me dramatically better progress-per-limit on a difficult, long-running autonomous engineering project. Not prettier answers. Not a nicer demo. More verified work completed per unit of usage. That’s what I actually needed.
Claude also completed working Approval → SafeAction → working-folder apply → post-verification → rollback in https://t.co/ENnxQp1Ifj, proved exactly-once behavior, fixed dependency/pytest/sandbox issues, and safely removed 118 obsolete workspaces. This is actual shipped engineering progress.
In roughly two days, Claude moved the same project through work that had been stuck for weeks:
• Telegram context/follow-ups fixed
• web research improved
• production fixes deployed
• canonical https://t.co/ehkFd01LLm → PASS
• runtime qualified + deployed
• large-repo verification passed
And that’s only part of it.
This is what Codex usage looked like.
My Analytics currently shows: Ashitaka - 184.75% of a weekly limit “аш��така” - 77.18%
And the latest fresh weekly allowance was almost completely burned again.
That would be easier to accept if the project were moving proportionally. It wasn’t. @thsottiaux @OpenAIDevs
@CodexResets1 Not only is there no reset, but support has been ignoring me for more than a week. Two week‑long limits have been exhausted. Each of us worked for less than a day on a single project.🥵
@prd_008 Not only is there no reset, but support has been ignoring me for more than a week. Two week‑long limits have been exhausted. Each of us worked for less than a day on a single project.🥵
@melvindvivas Not only is there no reset, but support has been ignoring me for more than a week. Two week‑long limits have been exhausted. Each of us worked for less than a day on a single project.🥵