Bot: @claude_popusk_bot
Support: @claude_popusk_support
Status: https://status.claude-popusk.shop
Lolzteam: https://lolz.team/threads/9981928
Claude Popusk ⚡️
Post #47
1.99K
🎭 Temporary pause
On August 7 our main upstream went down. We wanted to hit pause right away, but a lot of you asked us not to stop — so we went looking for alternatives.
The first backup upstream lasted about a day: the models were genuine, but there was no proper caching or reasoning control (we had to implement cache on our side). Under load it died quickly. We went back into maintenance and switched to another one.
The second upstream was worse. First came a wave of errors, then complaints about the models “lowered” intelligence. We checked ourselves: on paper the scores looked high (including on public benchmarks like LMSpeed), but in real use the outputs felt off — either model substitution (
We could go find yet another upstream. We won’t.
Since day one, Claude Popusk had one job — ship the highest-quality API we can. After losing a sane primary upstream, we can no longer guarantee that. Swapping one bad option for an even worse one is not our path.
So we’re temporarily pausing the new models.
What this means in practice:
—
— Some older models (like
— The bot, API, and web chat keep running — just without the new models.
— Balances don’t burn: you can keep spending them on the older models that remain.
Need a refund — message support. We’ll return your unused balance.
Personally: I’m tired of carrying everything alone — support, bugs, infra, reputation, and keeping the service honest. I don’t have the bandwidth to run it at the same quality right now, and I refuse to run it worse.
This is not a full shutdown. As soon as we can bring back our main (private) upstream, we’ll post here. I can’t give any timelines.
This channel will stay quiet. If you want to hear when real quality is back — just stay around.
Thank you for being with us this whole time. ❤️
On August 7 our main upstream went down. We wanted to hit pause right away, but a lot of you asked us not to stop — so we went looking for alternatives.
The first backup upstream lasted about a day: the models were genuine, but there was no proper caching or reasoning control (we had to implement cache on our side). Under load it died quickly. We went back into maintenance and switched to another one.
The second upstream was worse. First came a wave of errors, then complaints about the models “lowered” intelligence. We checked ourselves: on paper the scores looked high (including on public benchmarks like LMSpeed), but in real use the outputs felt off — either model substitution (
Fable 5 quietly served as Sonnet 5 / Opus 4.6), or extra system instructions injected on top. By now this upstream answers every other request and often returns straight-up garbage.We could go find yet another upstream. We won’t.
Since day one, Claude Popusk had one job — ship the highest-quality API we can. After losing a sane primary upstream, we can no longer guarantee that. Swapping one bad option for an even worse one is not our path.
So we’re temporarily pausing the new models.
What this means in practice:
—
Fable 5, Opus 5, GPT 5.6 and other new models are no longer available.— Some older models (like
GPT 4o) still work.— The bot, API, and web chat keep running — just without the new models.
— Balances don’t burn: you can keep spending them on the older models that remain.
Need a refund — message support. We’ll return your unused balance.
Personally: I’m tired of carrying everything alone — support, bugs, infra, reputation, and keeping the service honest. I don’t have the bandwidth to run it at the same quality right now, and I refuse to run it worse.
This is not a full shutdown. As soon as we can bring back our main (private) upstream, we’ll post here. I can’t give any timelines.
This channel will stay quiet. If you want to hear when real quality is back — just stay around.
Thank you for being with us this whole time. ❤️
- 👀 25

