gpt-6-astra is at capacity.. definetely openai are coocking 6.1 astra . today tomorrow or thursdai?

Dear tibo it is eod pt ptsd

world models are here !

No self hosted infra without k3s Kubernetes. Maintenance is the bottle neck: How iam solving maintenance automation? 1- obviously native cron jobs 2-promotheus + thanos signals 3- restate / temporal workflows 4-escalation to high reasoning ai agent ( gpt 6 astra or fable5.1) 5- escalation to me 6- self improvement loops ( produce new durable rules for the stateful workflows) 7-jev classifier in the middle My factory Is V1.5 and self building to 2.0. Maybe next move adopt @thorstenball Progressive swe engineering philosophy regarding what i call « swe bureaucracy » of PRs. But the practical alternative may come from anthropic scion , or from amp team , lets see ))

was opus 5.5 nerfed or it is because iam used to work with fable5.1?

Gpt 6.1 Luna Ultra-fast 👀

Opus5.5 needs a big brother fable5.5

Self hosted ai ? Yes of course more cpu for too cheap to meter intelligence, real cap is harware cpu memory not gpu . Running 64-128 luna ultra workers in my software factory with self hosted ci , 128 cores cpu 256 ram are not enough !

Tibo was silent for over 24h , maybe for the first time , is that big cooking? @thsottiaux

Do you know that chatgpt gas the best voice transcriber that works on every app where cursor blinks , unlimited free , no codex usage, highest accuracy . Like whisperflow with hot button and 95+% Accuracy that self improves with the context u building up across time.

Wake up tibo ! @thsottiaux

Sorry haiku5.5 , Luna 6 Won !

"O" My !!

Gpt6 Astra billing today is completely crazy, eaten 25% of usage from pro lite in under 1h, And 25% of pro200 in ( under 2h) . For same session that yesterday was using less than 10% on same tasks. Using codex cli . Only. Yesterday chatgpt codex app and codex cli .

Astra feels Sol , even on ultra still does lazy stupid mistakes! Was it nerfed before devday? Or opus5.5 effect?

If gpt6-astra was as fast as opus or 2x ? And similaraly priced as sol . Welcome to gpt6.1 On dev day. Intellegence of astra , speed of spark and price of sol.

Openai always ON , must be just enabling event driven codex wake ups ( instead of static sheduled tasks), also if codex can be programatically called over any mcp api hook will be huge unlock.

Skill distillation is hybrid approach. Take it with lots of grain of salt. It is like building software .

Gpt6 pro users be careful with your pro model usage, it is no longer the old fair unlimited. It is now hard capped ( at least some transparency) but very disappointing limit of 200 messages , yes 200 per WeeK !! Not per Day !! Gpt5.6 (underperform gpt6 extra high) has more usage, 170 per day. But unusable . When will openai make gpt6 pro default pro model at 170 messages per day as was 5.6? @OpenAI we need pro model usage back to 170+ per day !! You are damaging researcher, Iam competing for kaggle arc ag prize hackaton , but run out of pro messages hours ago . @OpenAIDevs @thsottiaux @gdb @sama

Sorry openai , real gpt6 is astra , sol looks like a renbanded 5.6 with no jacket