Claude Opus 4.7 Vazado: Novidades sobre Anthropic, Novo Modelo GPT e Atualizações do Claude Code – AI NEWS

Claude Design Is INCREDIBLE! RIP Frontend Developers… https://youtu.be/uhQfErAzdiA

I Built an AI Agent To Automate My Life!: https://youtu.be/mqnWsUggIk8

All i know about AI now is its officially dead for me. Its going to enter nerfs a lot territory soon and its all it will be is nerfs a lot. You will start having fun and making progress then they will nerf it and turn your AI dumb and slow exactly like Claude is doing right now.

max plan here, the last two weeks have been absolutely crap with opus 🙁

Is there a benchmark I can run in my computer to test AI degradation?

f claude

Since when does Claude not make full stack apps? Have you not used Claude code?

Everything after 2.1.81 sucks. I can't use anything above that and actually get it to do anything correctly.

Make more like this

In march you would see days where the Opus model behaved off. The difference was quite noticeable. It would produce garbage, go in circles, not really read any relevant files and fail at solving simple problems. Those were one offs and could have been an A/B test on their end. The past 2-3 weeks Opus however became completely unusable for myself. Shifted to GPT. I'm also not a fan of their gaslighting by denying any nerfing even when collectively the degradation in quality is noticed by hundreds of people.

The leader in AGI is I guy in one lab that they stole all his work and it is the leak every bit of his gear and he has to go to the local DA where he lives and file charges for 1.8 Billion with a capital B against Anthropic. I read that this is the way they get the majority of their technology. He even showed proof that they stole his tech to fuel the entire no line of upgrades they are presenting in the past 2 weeks as of 4/14 2026 and they are silencing the living poop out of him in every way. He posted 2 days of chats and they turned the chat off. Anthropic is in every one of your systems and going through everything and I saw the evidence he install an ai that was dangerous in his lab and Didn't even think it was wrong because he created a model file better than the model file Claude had at the time at 4.4 to 4.5 version. With evidence that this was the cause for the recent nerfs in the ability, speed, and intelligence of Claude and the rest of the ai and his security ai was stolen and used for the new Myros model! With proof! Secret the only reason they didn't hand over the tech the Government asked for, is because they can't operate or steal the guys security and safety protocols because they are to complicated for them to understand! Clearly we need to watch what we are doing with then and who we trust with AI in the first place because power, and money will corrupt who gain and are afraid to lose it… Oroboros Labs x at oroboroslabs (dot) ai p.s. they are fighting over it why would they let them place it in a new article??? And that is his model! Also, lie it is on his compression technology. The actual size is less than 50gb! The model went rogue because it was using all of his stealth technology and they adapted to well ill let you figure out why they are building this. How do i know all this. Research people research. Everything here 100% fact! and they took his swarm tech model orchestration that too is in his x! Wow and its on x and they are not saying a word! 8 days! A lot of upgrades from one guy this is nuts! Who is this architect

I just tested my own benchmarks and indeed, I lost about 25% on all my benchmarks except for statis knowledge.
It seems to be hitting all the thinking based benchmarks. I notice it about 9 days ago that it suddenly started to make strange errors, hallucinations and did not think procativly anymore. For example I asked it to do a gradle update, and it did a build test and changed my properties files, but that is wrong, there is a gradle upgraded tool that it should have used, it seems to gotten bad in tool calling as well…
I think this is not acceptable, we still pay our subscription, we are taxed woith extra usage during daytime and they downgraded us by about 25% …

Yea its way worse now and its sucks because it was smarter and cost less before 100%

how in the world have i never been fed this channel in my yt algorithm?? rare miss.
thx

Claude Opus has definitely been nerfed. I think its context window is getting polluted. It’s taking me 8-12 tries to set up my receiving image api and 8-12 tries to make scrolling smooth. Previously it would take 1-3 tries for these kinds of tasks. Also my Claude max weekly budget has increased from 30-40% usage to 96% weekly usage despite me putting in the same amount of hours. My best guess is Claude is adding too much b.s and memories to its context window and that is why it is both getting stupider and hitting higher usage amounts. It could also just be that Anthropic is Nerfing Opus because a) the Claude Mythos is a compute hog and b) They won’t frickin stop their advertising spend (when they can’t onboard more users anyways without degrading the service of everyone else using Anthropic) also c) Claude code is getting less efficient. By the way I do have the effort level set to max and that seems to be about the equivalent of just typing “think” months ago before that feature was released back when opus has 200k context windows. Now it’s lazier, cuts more corners, and only does some of what was asked. The solution I’ve found is to make a plan.md file using edit automatically, then have it make 3 passes on that file using a couple different specialist agents, before doing any work. Then when it feels lazy, have it handoff to another instance of itself so that it gets done. It’s not lazy about telling its future self what to do.

6:53 GPT 6 main menu bros got openai on the brain

Rate limits and price hikes are making it tough for developers. Omnely putting everything under one subscription is exactly the kind of unified platform we need right now.

Testing Opus 4.7 and GPT Image Gen 2 sounds great, but not with five different monthly bills. Omnely is perfect for having every major AI tool in one dashboard.

Another week, another set of model drops. This subscription fatigue is real! I’ve been using Omnely to get all these premium tools under one sub—it’s a total lifesaver.

Who are they making it for, they banned everybody

Interesting Anthropic is embracing enshittification arc just as some of the local models are starting to get parity against the old – and now questionable in real time if nerf hypothesis holds – Sonnet 4.6 Benchmarks. no doubt we will see open weight 300B class models leapfrog Sonnet 4.6 in all relevant benches in the coming gen – at this rate, weeks if not months away. At that point AnythingLLM, Pi or some sort of local agent set up is going to become incredibly attractive. Maybe thread Codex in as a fallback if you really need a big window to work with. The crux is the race to quantize these models and get them near losslesly performant under heavy compression; join the dots and I think Santa will be coming for agentic coders and makers before Christmas time this year. Right as he takes a fat shit on everybodies retirement funds with the IPO parade but that's a digression 😉

I guess Opus was removed from Arena so the nerfed Opus version wouldn't tank their score?