OpenAI is quietly cutting usage limits so they can launch a $500 plan.
This is a pricing scam built on opaque subscription limits.
Two months ago, burning through a $200 plan was hard.
Last year, the $20 tier was more than enough for almost everyone.
Now? A $200 tier fails after one day of real work. The $20 tier is just a free trial.
We need real consumer protection for AI subscriptions.
@Nlacombe_ What do you think about the card on my personal site? https://t.co/MWQVtwsrUO (not very OG invocation method tho but light/ dark modes required black and white 🥲)
@Eloise1041955 Alors juste attention car les prix affichés sont ceux avec abonnement annuel et si tu choisis pas dans quel datacenter tu veux le serveur et je sais que certains aiment pas car il faut donner beaucoup d’infos (en pratique je pense pas en avoir donné plus que chez OVH ou Scaleway)
If even half these rumors are true, the end of september is gonna be a hell of a month for AI 👀.
GPT-6 Sol: drop this week. Supposedly a smaller model from the Astra family. Name and specs still unconfirmed.
Opus 5.2: allegedly being tested in Claude Code under the Opus 5 name. Nothing official yet.
Grok 4.7: delayed after Elon said on September 11 it needed “a few more days to cook.” Could finally land this week.
Meanwhile, Kimi K2.8 Preview is already live in Kimi Code. Moonshot says it’s close to K3, with more efficient thinking.
Then there’s GLM 5.5, rumored for later this month. Open weights, 3T+ parameters, 1M context, supposedly taking on GPT-6 Astra and Mythos. Huge claims, but no confirmed specs or benchmarks yet.
Astra-level performance with open weights? That’s the one I’m most curious about.
Let’s see what actually ships.
My career adviser said my CV needed a more human side. So I spent a week building a site about the projects, people and experiences that shaped me.
Does it help you get to know me? I'd love your feedback.
It is something I tried about a month ago, using AI to rewrite 30+ microservices from @java (@springboot v4) to @rustlang. My main concern was RAM. Results on memory were truly amazing: from ~600MB usage for average service to ~6MB but there was a flaw. Build times increased up to 12 minutes for services which made it too hard for me to continue building at a reasonable pace not even mentioning the non mature tools it relies on. Solution for me was switching back to Java (but using @QuarkusIO this time) with multiple optimizations (and now using @OpenJ9 using ~80MB of RAM). Quarkus also offers native compilation but it sounds not that useful to me at his point and would cost more build time (tho I can chose my dev pods running on OpenJ9 and prod ones to be compiled to native executables using @graalvm so I still have room to improvement)
Et Apple Reference Image (une feature qui de souvenir signe les « pixels » de ton image pour qu’on puisse certifier qu’il étaient sur la photo originale) et j’ai un peu de mal à comprendre pourquoi vu que ça me semble aligné avec leur combat. (Annoncé pendant le apple event du 9 septembre)
https://t.co/uf81hgPgca
Ce soir c’est IOS27 !
Et si vous êtes Européen, voici ce que vous n’aurez pas : (déroulez, la liste est longue)
• Nouveau Siri conversationnel
• Contexte personnel (Messages, Mail, Photos…)
• Compréhension de l’écran
• Actions entre plusieurs apps
• Siri dans l’app Appareil photo
• Visual Intelligence avancée
• Write with Siri
• Style d’écriture personnalisé
• Nouvelle app Siri
• Nouvelle interface via Dynamic Island
• Nouvelles voix plus naturelles
• Siri AI sur Apple Watch, CarPlay et AirPods