Loading video...
Video Failed to Load
Opus 5.5 is proof that the world does actually need smarter models.
193,332 views • 12 days ago •via X (Twitter)
29 Comments

Coders don't want smarter models because they still want to feel useful. They're terrified that normies will software mog them without even knowing what a database is. One day people who can't write a single line of code will be able to generate a GTA 6 level game... with one prompt. Times almost up nerds

“the world needs smarter models” is a bold claim from a timeline that still asks them how many r’s are in strawberry 🐈⬛

the world needs cheaper smart models more than smarter ones. my bill says so

every time i think models are getting good enough, a new release shows me how much further they can go.

1. Not everyone has time and money to build their own software 2. Enterprise solutions usually require strict licenses, e.g. SOC 2 Compliant 3. The benchmark must be carefully reviewed by human, not just created by Opus 5.5 and reviewed by Sol 6.1

Few correction glanced over the transcript didn't watch the vid SM64 was decompiled in 2019 before llms were mainstream or useful for any task all the decomps you see online are just grabbing byte code and bundling with emulators they take advantage of previous dev work like yuzu

Fable 5.5 is the next.

the gap shows on the messy tasks, opus 5.5 just keeps going where others stall

but it is really genuinely hard to imagine a model better than Opus 5.5, currently really nothing comes close...and doesn't even seem required honestly.

We still need smarter models 100% And honestly in my opinion not better models at coding, but rather more intelligent models like just general intelligence Not sure if that makes sense...

Theo, please consider running the benchmarks on opus 5.5 again. I bet it will score significantly lower today.

models get smarter every month and yet the cold emails in my inbox somehow keep getting dumber

Opus 5.5 felt like a downgrade in it output style

ok this video is gold

Agreed. Opus 5.5 fixed stuff in one pass that I'd been going back and forth on with smaller models for days

dumb models need smart users, i guess

I have loved seeing how ambitious and creative people are

For real if they can just keep updating this model, no ai model that will be released in the future will be better that it.

The gains are showing up in places benchmarks miss. I fed Claude Haiku 20 subtle false premises (Eiffel Tower opened in 1899, ChatGPT launched in 2021). Over 5 runs it caught all 100. Two years ago, models just played along.

Your hair is fucking horrendous

@grok summarize

smarter matters most in the boring part: reading a codebase it didn't write and not breaking something two files away

You can’t anticipate what you would do with more intelligence. Glad you came around.

There is much to optimize in the current world. More insane, is that the more we optimize the current software, the less compute the world will use in the future, and the loop to building stuff will become nearly real-time.

smarter models means easier access to AI for all. that's really what we should be striving for as a civilization

time to let it write code I won't understand

Smarter models, same me hitting approve on every command it asks about

Bruh 31 minutes???

whoever told you to remove the mustache and do your hair substantially better is the real goat
