ALL OF THIS WAS BUILT WITH OPUS 5.5 I've upgraded RONIN's combat system like 100x. Blocking and dodging KI points Wall jumps Finishing And SO MUCH MORE. I've spent over 80 hours working with Opus 5.5. I'll be showing everything I've built very soon... You haven’t even seen 1%.

Fable 5.5 will NOT be today; it’s quite unlikely they’ll post filler and launch their monstrous model two hours later. We’ll have to wait until next Tuesday.

I see a lot of people complaining about the intelligence of OpenAI’s models, but these days, taste is valued much more highly for 99% of users, and I don’t see anyone asking OpenAI for that. Please, guys, if you’re going to release a good model, MAKE IT EXTREMELY CREATIVE and give it real taste.

OpenAI releasing a new Terra model has to be a lie... THIS WOULD LITERALLY BE GOING F**KED.

▶️Tip for Claude Code: If you want to save tokens while using Opus 5.5, set "/Auto-compact 512k" Trust me, it makes an ABYSMAL difference in your usage.

Claude max x20 is literally infinite, those of you who already know me know I don't hype things for no reason, I swear it's insane, I had opus 5.5 running on High for 14 hours (YES, 14 HOURS) alongside 6 continuous opus 5.5 medium subagents, and it only used 21% of my quota.

After using Claude for a whole month to create video games, I can tell you this: OPUS 5.5 MEDIUM is ALL you need.

My deepest reflection on the situation with OpenAI (please, if you read the whole thing, I hope you’ll share your thoughts on it too. I really want to hear what you think my frienddd): Lately, I’ve seen far too many people arguing that OpenAI owes us absolutely nothing, that we should be grateful we get to use its technology, and that complaining about its decisions is a sign of arrogance. Well, honestly, I find it worrying that we’ve so normalized defending big companies that we even question consumers’ right to demand improvements. Imagine you live in an apartment building. You pay monthly for your home and for the upkeep of the shared facilities. When you moved in, there was a beautiful garden, working elevators, and running water on every floor. Over time, management stops looking after the garden, the elevators start breaking down, and water barely reaches the upper floors. When the neighbors start complaining, someone shows up and says: “You should be grateful you have a roof over your head, and management owes you nothing.” Excuse me? I’m not demanding that they build me a swimming pool. I’m asking them not to let the things I’ve been paying for every month deteriorate, and that’s exactly what happens with a lot of AI companies (right now it’s OpenAI, but we know that other times it’s been Claude, or whoever the fuck. I don’t care). You can spend months getting used to certain features, integrating them into your work, and enjoying a particular experience, until an update changes everything you valued. Or, in the worst case for me, they all keep evolving and you’re left without any improvements. Of course companies have the right to evolve their products. But evolving doesn’t always mean improving, and an update doesn’t automatically become positive just because it’s newer. In OpenAI’s case, I’m not demanding that they keep every model, voice, or feature forever. I’m not saying my personal preferences should come before those of millions of users either. What I’m saying is that listening to your customers doesn’t mean blindly obeying them. It means recognizing that their experiences and criticism have value. If thousands of people agree that a feature has gotten worse, the reasonable thing is to investigate what’s going on, not treat their complaints as if they were being ungrateful. And there’s something that seems especially absurd to me... many of the people saying today that we should keep quiet enjoy products that improved precisely because other consumers complained before them. And yes, I can always cancel my subscription and go to a competitor. But reducing any business relationship to “if you don’t like it, leave” is a terribly shallow view of how a market should work. I deeply admire what OpenAI has achieved. I consider its technology revolutionary, and that’s precisely why I use it and pay for it (though I’ve now canceled and switched to Claude, but sometimes it’s the same with them). I don’t think OpenAI owes me obedience, or that it has to develop a product exclusively for me, but I also don’t think those challenges should become a universal excuse to dismiss any criticism. Paying doesn’t give me the right to demand just anything. But it does give me the right to question the quality of the service I receive. And when a community has been pointing out the same problems for months, the least I expect from a company I support financially is that it listens, considers that criticism, and, when possible, explains its decisions. Because, going back to the apartment building example, we’re not asking them to give us the building. We’re asking them to take care of what we pay for. Anyway, I know this may be a bit of a slog, but honestly, I don’t understand when defending something as basic as getting better service for what you pay for started to feel like something you have to keep asking for, and if you do, some people look at you funny, thinking nobody owes you anything. You piss me off so much, I’m just going to say it plainly.

⚠️ CODEX and CLAUDE CODE USERS: I’m going to give you probably the best piece of advice you’ll ever hear, even if you don’t see its value. In your agent’s instructions, write "If you need to delete any files, review your command twice before running it, regardless of whether the auto-review subagent is reviewing it in the background. Also, files will always be sent to the recycle bin (and you will never skip this), so you will NEVER permanently delete a file, and if the recycle bin is full, you will stop and ask me to empty it myself before continuing." Believe me, I’m going to make life easier for a lot of you and probably spare many of you a lot of grief. I’m not joking at all.

I'm sorry @OpenAI @thsottiaux, I love OpenAI, but Opus 5.5 surpasses absolutely everything in creativity and taste, besides being faster and more powerful... no hate.

All of this is really useful! I recommend checking it out :)

Veo a muchos listillos criticando al Xokas de como usa la IA por aquí, antes que nada voy a dejar claro que no sigo continuamente su contenido, apenas veo a streamers, pero de inteligencia artificial sí que os puedo hablar y os voy a aclarar varias cosillas. 1. Un MCP es básicamente el "puente" que te permite que un modelo de IA controle un programa mediante scripts, realmente lo está haciendo bien y así es como debería hacerse, todos aquellos que os reís, no tenéis ni puta idea jaja. 2. El usar GPT 6 Astra aunque sea para conectar únicamente el MCP a Davinci tampoco está tan mal, ya que si está cambiando entre modelos baratos (Como 6 luna o 6.1 Sol ligero) para enviar mensajes cortos y luego cambiar a Astra para tareas potentes, gastaría muchos más tokens y límites ya que está literalmente haciendo que cada nuevo cambio el modelo tenga que releer toda la conversación de nuevo y jode la caché y el contexto por completo. 3. Ahora mismo GPT 6.1 Sol es prácticamente a nivel de astra y mucho más eficiente, podría usarlo sí, pero os aseguro que puede permitirse esto y más, incluso 5 planes de 500 para quemarlo con /ultrafast. 4. Los que comentáis que usáis "Ultra" creo que no sabéis que "muy alto" o "max" usa incluso más razonamiento que ultra, ultra está pensado para subagentes, lo que hace que el modelo principal solo orqueste y no razone tanto, si no que delegue sus tareas en subagentes más pequeños, además, os revienta la cuota de uso. Podéis investigar sobre el "Juice" en Codex para informaros de esto. 5. Estoy 100% seguro que ninguno usáis prompts estructurados, de hecho a día de hoy el prompt engineering está muriendo y los modelos son completamente capaces de comprenderos sin necesidad de estructurar un prompt, es mejor hacerlo, sí, pero el 99% de las personas que decís que lo hacéis o os reís, no lo hacen jajaja (y me incluyo). Soy el primero que ve cosas así y puede darme o no coraje, pero igualmente recordad que no todos nacisteis sabiendo y probablemente ni siquiera el 90% de los que critican saben lo que es Codex y solo usan ChatGPT para que les diga como freír un huevo o corregir errores de ortografía de una imagen subida. Ya me gustaría veros a vosotros usar Codex 🤣

Nano Banana 3 PRO has been spotted in the internal code. If this really does end up arriving soon, I’m definitely sure it’s going to be very difficult to beat them when it comes to image generation. They’ve been the GOATs, without a doubt.

What the hell? They really just lied in the benchmarks, there's no comparison. Luna medium would do this halfway better.

MUSE WONDER BENCHMARKS What the hell just happened? Meta just cooked everyone. Let's hope these benchmarks reflect reality, but I'm absolutely impressed.

Was Opus 5.5 nerfed? Lately, I've noticed Opus 5.5 being much less proactive. It also gets stuck thinking and repeatedly running tests and making changes on tasks that should be much faster, only to then deliver work that's noticeably worse than usual...

Look at this, nice for ultracode users! So cheap :)