Video yükleniyor...
Video Yüklenemedi
I have feelings about Opus 4.7.
149,677 görüntüleme • 7 ay önce •via X (Twitter)
37 Yorum

who edited it?

Me lol

but it was fun right?

Yeah I miss editing a ton, always fun to dive back in

i do have some data here showing very concrete improvement compared to 4.6 in its ability to invoke the right skills for the right tasks - still not as good as gpt 5.4 in codex though

Honestly, Anthropic did a big favour for OpenAI with this release

man we're eating good this week. channel content has been dope

I WAS WAITING SO IMPATIENTLY FOR THIS

It mostly likely all anthropic models based of Mythos like 4.7 will keep having this problem

I just finished watching the open source video and @theo drops another banger.

Watching it now, I agree about the classifiers being ridiculous. Much love showing how 'safe' it is!

I was wondering if I’m the only one thinks 4.7 is dumber than 4.6

I wanted to see the last part, please take care of your cables next time. Good edit though.

Safety measures at its finest.. the model now refuses its own job 😂

theo drinking beer before mythos

Do you ever sleep Theo

The final 5 mins were savage.

I hit my 20x Claude Code usage cap while testing out research subagent fleets on Opus 4.7 Max. After my usage cap reset at 10pm, I told 4.7 to continue working. This is what it hit me with. Yeaaa this model sure is something, not sure how much of this is a harness issue tho

thank you Theo for editing this video while being drunk

the patron saint of anthropic haters return. hail saint theo.

was waiting for it

Diving in right away

Opus 4.7 blocked my incredibly dangerous... Claude code plugin release. Sonnet 4.6 was perfectly happy to download the CC0 license and write a change log though. If curious, the 'identity' it's talking about is a GitHub org.

You're drinking the wrong beer!

Just watch the entire video and agree with your findings. Sad to see the regression and that it seems that new releases will follow this downward trend. Also, it seems that although the harness is the main problem, some of the quality issues poisoned the model also.

i wrote about this:

Use it in a better harness man… works with Claude oauth as of April 17

of course you do

Why does it suck —> because they aren’t using the same tools we use. Why? Their culture and values hold that intelligence gets dangerous when distributed. This is culture and can’t be fixed. Anthropic is going to lose to OpenAI and weirdly xAI. Because of culture.

I also think it’s the weirdest model ever released, and I question why they would release this.

Totally agree with this. Every release is rushed, bad implementations and bugs where they just shouldn’t be. If you ain’t using your own shit though…

@grok what does theo think here. NO. what does he feel!

yeah. release cadence is just outrunning the actual capability jumps now.

it referenced a march 2026 paper i know it can be web search or whatever but im excited about the future.

I feel your frustration. I have the exact same, except yours were without much cursing at it. lol 😂 I didn't want to give OpenAI a try, but now I kinda feel like I have no other choice left. I have tolerated Opus 4.6 too long to then get my health affected by Opus 4.7 due to it's stupidities.

"i have feelings about 4.7" -- who doesn't :D j/k, was waiting for your vid, ty

@grok resume the take away keyPoint
Benzer Videolar
I peered into Claude's soul, and I have feelings about it.
Theo - t3.gg
52,483 görüntüleme • 9 ay önce
