\
  The most prestigious law school admissions discussion board in the world.
BackRefresh Options Favorite

new claude model is a "step change" in AI capabilities

it's going to be 180. A draft blog post that was availabl...
cracking filthy boistinker fortuitous meteor
  03/28/26
How can you still be falling for this shit man They're ju...
Flickering big-titted abode
  03/28/26
...
vivacious gay wizard theater
  03/28/26
The level of lying and fraud just keeps increasing. It's off...
Flickering big-titted abode
  03/28/26
wow
Dagoth Ur
  04/05/26
How is it a lie when the latest models already code and do a...
Odious infuriating chapel digit ratio
  03/28/26
Glorified autocomplete brah…you’re being Dunnin...
mog
  04/06/26
I don’t care what a scammer like Altman says. The mode...
cracking filthy boistinker fortuitous meteor
  03/29/26
This "Mythos" shit already got called out as bulls...
Flickering big-titted abode
  03/29/26
...
Wonderful fantasy-prone new version pocket flask
  03/29/26
...
British Upgrade
  04/05/26
How the hell are you hitting the quota limits on the api? I ...
in the naked
  04/05/26
...
glassy base
  03/29/26
It's gotten dramatically better at legal research during the...
German pumo
  04/05/26
Isn’t this the one that got Hacked and is open source ...
cowgod
  04/05/26
claude code did. not the ai model/weights itself. wasn't hac...
in the naked
  04/06/26
But there are parameters Encoded in Claude code which give a...
cowgod
  04/06/26
not the neural model itself. the cli/agent harness system ha...
in the naked
  04/06/26
This is Huge
cowgod
  04/06/26
meh. whats amazing is that these idiots leaked so much in su...
in the naked
  04/06/26
I was hoping there’s be a way to host the chat it loca...
cowgod
  04/06/26


Poast new message in this thread



Reply Favorite

Date: March 28th, 2026 2:54 PM
Author: cracking filthy boistinker fortuitous meteor

it's going to be 180.

A draft blog post that was available in an unsecured and publicly searchable data store prior to Thursday evening said the new model is called Claude Mythos and that the company believes it poses unprecedented cybersecurity risks.

“Compared to our previous best model, Claude Opus 4.6, Capybara gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others,” the company said in the blog."

https://fortune.com/2026/03/26/anthropic-says-testing-mythos-powerful-new-ai-model-after-data-leak-reveals-its-existence-step-change-in-capabilities/

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49776615)



Reply Favorite

Date: March 28th, 2026 3:02 PM
Author: Flickering big-titted abode

How can you still be falling for this shit man

They're just lying. They're always lying. They've been continuously lying nonstop for years. Every single thing they say is a lie. Every claim about new capabilities is a lie. It's just all lies

Dario and Sam have both announced "AGI" like 50 times each at this point. The entire "space" is a pump and dump scam

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49776641)



Reply Favorite

Date: March 28th, 2026 3:03 PM
Author: vivacious gay wizard theater



(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49776643)



Reply Favorite

Date: March 28th, 2026 3:10 PM
Author: Flickering big-titted abode

The level of lying and fraud just keeps increasing. It's off the fucking charts at this point

This new fraud "step change" model is being released with lies about its increased capabilities so that they can massively increase their pricing. Then everyone else will release their own "step change" models and increase their own pricing. The cabal benefits

That's all that any of this is. It's just a fucking scam industry and this shit is so played out and tiresome already



(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49776656)



Reply Favorite

Date: April 5th, 2026 11:57 PM
Author: Dagoth Ur

wow

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796797)



Reply Favorite

Date: March 28th, 2026 3:08 PM
Author: Odious infuriating chapel digit ratio

How is it a lie when the latest models already code and do advanced math and physics better than any human?

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49776652)



Reply Favorite

Date: April 6th, 2026 8:57 AM
Author: mog

Glorified autocomplete brah…you’re being Dunning Kruger mogged. May want to sit this one out…

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49797124)



Reply Favorite

Date: March 29th, 2026 12:49 PM
Author: cracking filthy boistinker fortuitous meteor

I don’t care what a scammer like Altman says. The models have become much better over the last year. Agentic coding actually works now and it appears the models for it are highly likely to become much better soon

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49778346)



Reply Favorite

Date: March 29th, 2026 2:43 PM
Author: Flickering big-titted abode

This "Mythos" shit already got called out as bullshit flame btw

It's all so tiresome

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49778564)



Reply Favorite

Date: March 29th, 2026 2:44 PM
Author: Wonderful fantasy-prone new version pocket flask



(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49778568)



Reply Favorite

Date: April 5th, 2026 11:44 PM
Author: British Upgrade



(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796784)



Reply Favorite

Date: April 5th, 2026 11:55 PM
Author: in the naked

How the hell are you hitting the quota limits on the api? I have never used the anthropic api directly I have used it through azure and azure has TPM limits idk what they add up to. But on open AI the api has tiers that go up to something like $200K worth a month? How would you even hit that? I remember for a while I was using personal codex credits and I would max it out every week (it refreshed weekly) and I was pushing like almost 10K lines of code from it to github per week on that before wasting money on api credits.

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796794)



Reply Favorite

Date: March 29th, 2026 12:59 PM
Author: glassy base



(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49778356)



Reply Favorite

Date: April 5th, 2026 11:49 PM
Author: German pumo

It's gotten dramatically better at legal research during the year and a half I've been using it.

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796786)



Reply Favorite

Date: April 5th, 2026 11:59 PM
Author: cowgod

Isn’t this the one that got Hacked and is open source now

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796802)



Reply Favorite

Date: April 6th, 2026 12:00 AM
Author: in the naked

claude code did. not the ai model/weights itself. wasn't hacked some dumbass who worked there made it public on accident

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796803)



Reply Favorite

Date: April 6th, 2026 12:02 AM
Author: cowgod

But there are parameters Encoded in Claude code which give away the Engineering of the model imho. This is my opinion as a game Engineer and not a fact.

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796805)



Reply Favorite

Date: April 6th, 2026 12:05 AM
Author: in the naked

not the neural model itself. the cli/agent harness system has a lot to do with claude code performance though and thats leaked.

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796811)



Reply Favorite

Date: April 6th, 2026 12:27 AM
Author: cowgod

This is Huge

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796831)



Reply Favorite

Date: April 6th, 2026 12:46 AM
Author: in the naked

meh. whats amazing is that these idiots leaked so much in such a short time. first they leaked all the articles talking about the new upcoming model. then the full claude code codebase which had all kinds of references to unreleased features in it

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796848)



Reply Favorite

Date: April 6th, 2026 7:05 AM
Author: cowgod

I was hoping there’s be a way to host the chat it locally and jailbreak it for anti semitism

(http://www.autoadmit.com/thread.php?thread_id=5850944&forum_id=2...#49796994)